A task decomposition and federal distribution method and system based on a multi-level knowledge graph

Abstract

The application relates to the technical field of artificial intelligence and organization management, and discloses a task disintegration and federal distribution method and system based on a multistage knowledge graph, which adopts a semantic analysis method to convert unstructured to-be-processed documents into structured task carriers for task distribution, and synchronously extracts key information; a double-layer knowledge graph of an organization unit and a member individual dimension is constructed, and correlation is realized through a cross-dimension correlation algorithm; in combination with a privacy protection and federal learning mechanism, the correlation degree of a task and the organization unit is calculated to match the organization unit, each organization unit quantizes the member capability locally, and after encryption, the data is aggregated to generate a task list; according to member task execution feedback data, the task distribution weight of the member is dynamically adjusted, and the overall load balance of the system is guaranteed. The application forms a closed loop of "analysis-modeling-distribution-optimization", significantly improves the task distribution accuracy and the system response efficiency, and adapts to the efficient cooperation demand of cross-department report tasks of enterprises and institutions.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and organizational management technology, specifically to a method and system for task decomposition and federated allocation based on a multi-level knowledge graph. Background Technology

[0002] When enterprises and institutions collaborate across departments to complete reporting tasks such as annual reports of listed companies and enterprise work reports of the National Audit Office, existing task decomposition and allocation technologies have multiple pain points and are unable to meet the needs of efficient, secure, and dynamic collaboration. These pain points are mainly reflected in the following aspects:

[0003] I. Regarding task breakdown, the traditional model relies entirely on staff experience, requiring chapter-by-chapter report analysis, clear task boundaries, and departmental responsibilities. This process is time-consuming and highly dependent on staff capabilities. Experienced staff need 2-5 days to break down a single report, while novices require an additional 1-2 days of training. Furthermore, errors and omissions are prone to occur, necessitating rework. Simultaneously, manual breakdown lacks standardized procedures, leading to inconsistencies in task assignment among different personnel and hindering the development of a standardized workflow. When the number of reports surges or the content becomes complex, time costs increase linearly, easily delaying the overall compilation process.

[0004] Second, regarding data management, existing employee competency data (such as skill proficiency and task completion quality) from various departments are scattered and stored in independent systems, forming data silos. When assigning tasks across departments, the coordinating entity has difficulty accessing data from other departments, which can easily lead to a mismatch between tasks and employee capabilities. Furthermore, existing technologies lack secure sharing solutions; direct sharing can easily leak privacy and core data, while not sharing means relying solely on subjective recommendations, creating a dilemma between sharing and efficiency.

[0005] Third, regarding task allocation mechanisms, mainstream systems (such as SAP's task module) have static flaws in allocating tasks according to preset rules. They cannot respond to load changes, leading to uneven workloads and wasted resources. Furthermore, they only support single-level allocation and cannot achieve multi-level linkage such as "report breakdown → department matching → employee matching," requiring manual intervention. More importantly, the existing mechanism does not consider privacy protection, transmitting and storing sensitive information in plaintext, which is prone to leakage.

[0006] In summary, the existing technologies have multiple shortcomings in terms of task decomposition efficiency, data sharing security, dynamic allocation adaptability, and privacy protection, which have seriously restricted the efficiency and quality of cross-departmental collaborative tasks and failed to meet the organization's needs for intelligent, secure, and efficient task processing. An innovative technical solution is urgently needed to address these pain points. Summary of the Invention

[0007] In view of this, in order to address the multiple technical problems of existing technologies in cross-departmental collaborative tasks, such as the efficiency of task decomposition, the security of data sharing, the adaptability of dynamic allocation, and the protection of privacy, this invention provides a task decomposition and federated allocation method and system based on a multi-level knowledge graph. This method enables intelligent and accurate decomposition of report content, dynamic matching of departmental responsibilities and employee capabilities, and real-time optimization of task allocation paths without centralized acquisition of privacy data from various departments.

[0008] In a first aspect, the present invention provides a method for task decomposition and federated allocation based on a multi-level knowledge graph, including:

[0009] Semantic parsing is used to parse unstructured documents that require cross-departmental collaboration, extract key information related to task allocation from the documents, and transform the unstructured documents into structured task carriers that can be used for task allocation.

[0010] A two-layer knowledge graph covering organizational unit and individual member dimensions is constructed. Cross-dimensional association algorithms are used to model cross-level relationships, and the organizational unit dimension graph and the individual member dimension graph are mapped to a unified semantic space to associate the two types of graphs.

[0011] Task allocation is carried out for organizational units and individual members based on a privacy protection mechanism combined with a federated learning mechanism;

[0012] Obtain feedback data on individual member tasks and dynamically adjust the task allocation weights of individual members to achieve overall system load balancing.

[0013] The task decomposition and federated allocation method based on a multi-level knowledge graph provided in this invention starts from the entire process of cross-departmental collaborative task processing. It uses semantic parsing to transform unstructured documents into structured task carriers, solving the problems of inefficiency and inconsistent standards in traditional manual decomposition, and providing accurate basis for subsequent allocation. The construction of a two-layer knowledge graph and cross-dimensional association breaks down data barriers between organizational units and individual members, compensating for the shortcomings of existing single-level allocation technologies. The combination of privacy protection and federated learning ensures data security while achieving cross-departmental collaborative allocation, resolving the dilemma of data silos and privacy leaks. Dynamic weight adjustment responds to load changes, avoiding resource waste. The overall system forms a closed loop of "parsing-modeling-allocation-optimization," significantly improving task allocation accuracy and system response efficiency, and adapting to the efficient collaboration needs of cross-departmental reporting tasks in enterprises and institutions.

[0014] In one optional implementation, the key information related to task allocation includes at least one of the following: document chapter structure, dedicated organizational unit identifier, and core task description terminology; the structured task carrier is a structured task tree, and when constructing the structured task tree, key fields and task nodes are extracted from the unstructured document to be processed, and the task tree structure is built based on the association between the key fields and task nodes.

[0015] This invention clearly defines the types of key information and the structured task carrier format, making the semantic parsing process more targeted and operable. The extracted key information, such as document chapter structure and unique organizational unit identifiers, can be directly used for rapid matching of subsequent tasks and organizational units, reducing manual identification costs. The structured task tree, built through the association of key fields and task nodes, has a clear hierarchy and complete information, intuitively presenting the task decomposition logic and facilitating accurate location of the corresponding organizational unit and individual member in the subsequent federated allocation process. Simultaneously, the unified task tree construction rules avoid discrepancies caused by different personnel parsing, forming a standardized decomposition process, further improving the efficiency and consistency of task decomposition, and laying a standardized data foundation for the entire allocation process.

[0016] In one optional implementation, the organizational unit dimension graph uses organizational units as nodes and the relationship between rights and responsibilities between units as edges, while the individual member dimension graph uses individual members as nodes and the collaborative relationship between individuals as edges; the relationship between rights and responsibilities between organizational units includes approval relationships and collaborative relationships, and the collaborative relationship between individual members uses historical collaboration frequency as a quantitative indicator.

[0017] The cross-dimensional association algorithm is the TransR algorithm. It is used to model cross-level relationships, specifically including: for entities h and t in the organizational unit dimension graph and the cross-level relationship vector r, and entities hr and tr in the member individual dimension graph and the relationship vector r, the entities are projected from the entity space to the relationship space using the relationship projection matrix Mr. The association relationship satisfies... ;

[0018] Among them, h r = h M r, t r = t M r Mr ∈ R k×d Let be the projection matrix corresponding to the relation vector r, k be the entity vector dimension, and d be the relation vector dimension.

[0019] This invention defines the node and edge attributes of a two-layer knowledge graph, enabling the graph to accurately map the responsibilities and authority of organizational units and the collaboration of individual members, providing high-quality data support for cross-level associations. Clearly defined approval and collaboration relationships between organizational units facilitate rapid task matching to corresponding responsible departments; quantifying the frequency of individual member collaboration allows for prioritizing well-coordinated employee combinations, improving execution efficiency. The TransR algorithm achieves a unified semantic space mapping between the two types of graphs through a projection matrix, effectively solving the problem of cross-level semantic misalignment, ensuring precise alignment between departmental responsibilities and employee capabilities, and overcoming the limitations of existing technologies that only provide single-level allocation and lack cross-level associations.

[0020] In one optional implementation, the step of using a privacy-preserving mechanism combined with a federated learning mechanism to allocate tasks to organizational units and individual members using key information related to task allocation includes:

[0021] Calculate the correlation degree between the responsibilities and powers of each task and organizational unit in the structured task carrier, and filter the matching organizational units based on the correlation degree;

[0022] Each organizational unit quantifies the individual capabilities of its members locally and encrypts the quantification results through a privacy protection mechanism.

[0023] The encrypted individual ability data of members from various organizational units are aggregated to generate the final task assignment list.

[0024] This invention integrates privacy protection and federated learning mechanisms into the entire task allocation process. First, it filters and matches organizational units through correlation calculations to ensure task allocation aligns with departmental responsibilities, avoiding allocation deviations caused by unclear responsibilities. Each organizational unit locally quantifies and encrypts member capabilities, ensuring the original capability data remains locally, effectively preventing data leakage and protecting departmental and member privacy. Federated aggregation of encrypted data generates a task list, eliminating the need for centralized acquisition of original data from each department. This breaks down data silos, enabling cross-departmental data collaboration while ensuring data security. This process balances allocation accuracy and data security, addressing the problems of unreliable allocation or insecure shared data in existing technologies, and improving the security and reliability of cross-departmental task allocation.

[0025] In one optional implementation, cosine similarity is used to calculate the correlation between the task and the responsibilities of the organizational unit, and a correlation threshold is set. When the calculated correlation is greater than the threshold, the task is determined to match the corresponding organizational unit.

[0026] For tasks with a unique organizational unit identifier, they are directly associated with the corresponding organizational unit.

[0027] The privacy protection mechanism is a differential privacy mechanism, which achieves privacy protection by adding Laplace noise to the quantification results of individual member abilities, as shown below:

[0028] F=E i +Laplace(0,β)

[0029] Where F is the encrypted individual member ability vector; E i Let E be the original capability vector of individual member i. Let Laplace(0,β) represent random noise with a Laplace distribution of position parameter 0 and scale parameter β, used to modify the original capability vector E. i Perform perturbation encryption.

[0030] This invention employs cosine similarity calculation to provide a quantitative standard for the correlation between tasks and organizational unit responsibilities. Combined with a correlation threshold, this makes organizational unit matching more objective and accurate, avoiding subjective judgment errors. Dedicated organizational unit identifiers directly link tasks, eliminating the correlation calculation step and further improving allocation efficiency. The differential privacy mechanism uses Laplace noise for encryption, protecting the privacy of individual member data while maintaining data availability. This ensures accurate employee matching based on encrypted data during federated aggregation, balancing privacy protection and allocation accuracy. It meets data security compliance requirements while guaranteeing accurate and efficient task allocation, solving the problem of insufficient privacy protection or excessive encryption leading to allocation failure in existing technologies.

[0031] In one optional implementation, obtaining feedback data on individual member task execution and dynamically adjusting the task allocation weights for individual members includes:

[0032] Obtain feedback data on individual member task execution, including task completion progress, task completion quality, and task processing time, to update the individual member's ability quantification results and task backlog.

[0033] The task allocation weights are dynamically adjusted based on the backlog, and the weights are updated using the following formula:

[0034]

[0035] Where n represents the total number of members in the current system who can be assigned tasks; Q i The number of tasks to be processed for member individual i; Q represents the total number of tasks to be processed by all individual members. avg The average number of tasks to be processed for all individual members; η represents the learning rate, which is dynamically adjusted based on system performance.

[0036] η t =η t-1 exp(-0.1 |Q avg - Qi |);

[0037] Where, η t η represents the learning rate at the current moment. t-1 η represents the learning rate at the previous time step, and η has preset upper and lower limits. exp() is an exponential function with the natural constant e as its base.

[0038] This invention explicitly focuses on feedback data encompassing task completion progress, quality, and time consumption, comprehensively reflecting the individual task execution status of team members. This provides a realistic and dynamic data basis for updating capability quantification results and task backlog, ensuring that subsequent weight adjustments align with members' actual capabilities and workload. The weight update formula based on backlog and dynamic learning rate reduces the weight of members with excessive workloads and increases the weight of members with insufficient workloads, effectively achieving system load balancing and avoiding the uneven workload and resource waste caused by static allocation in existing technologies. The exponential decay adjustment and upper / lower limit settings of the learning rate prevent excessive weight adjustments from causing system oscillations, ensuring a smooth adjustment process. Overall, this achieves dynamic optimization of task allocation, improving system resource utilization and task response speed, and adapting to scenarios with fluctuating task loads.

[0039] Secondly, this invention provides a task decomposition and federated allocation system based on a multi-level knowledge graph, the system comprising:

[0040] The document structuring module is used to parse unstructured documents that require cross-departmental collaboration using semantic parsing methods, extract key information related to task allocation from the documents, and transform the unstructured documents into structured task carriers that can be used for task allocation.

[0041] The two-layer knowledge graph construction module is used to construct a two-layer knowledge graph covering the organizational unit dimension and the individual member dimension. It uses cross-dimensional association algorithms to model cross-level relationships, maps the organizational unit dimension graph and the individual member dimension graph to a unified semantic space, and associates the two types of graphs.

[0042] The task allocation module is used to allocate tasks to organizational units and individual members based on a privacy protection mechanism combined with a federated learning mechanism.

[0043] The dynamic adjustment module is used to obtain feedback data on the tasks performed by individual members and dynamically adjust the task allocation weights of individual members in order to achieve overall system load balancing.

[0044] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the task decomposition and federated allocation method based on a multi-level knowledge graph as described in the first aspect or any corresponding embodiment.

[0045] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the task decomposition and federated allocation method based on a multi-level knowledge graph as described in the first aspect or any corresponding embodiment thereof.

[0046] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the task decomposition and federated allocation method based on a multi-level knowledge graph as described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0047] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art 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 from these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating the task decomposition and federated allocation method based on a multi-level knowledge graph according to an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram illustrating task allocation using various task processors according to an embodiment of the present invention;

[0050] Figure 3 This is a structural block diagram of a task decomposition and federated allocation system based on a multi-level knowledge graph according to an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] This embodiment provides a task decomposition and federated allocation method based on a multi-level knowledge graph. Figure 1 This is a flowchart of a task decomposition and federated allocation method based on a multi-level knowledge graph according to an embodiment of the present invention. It should be noted that the steps shown in the flowchart can be executed in a computer device such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here. Figure 1 As shown, the process includes the following steps:

[0054] Step S1: Using semantic parsing, unstructured documents that require cross-departmental collaboration are parsed to extract key information related to task allocation and transform them into structured task carriers that can be used for task allocation.

[0055] In this embodiment of the invention, the key information related to task allocation extracted from the document includes at least one of the following: document chapter structure, exclusive organizational unit identifier, and core task description terminology; the structured task carrier is a structured task tree. When constructing the structured task tree, key fields and task nodes are extracted from the unstructured document to be processed, and the task tree structure is built based on the relationship between the key fields and task nodes.

[0056] The document to be processed is the "2024 Corporate Work Report," stored in Word / PDF format. It includes chapters such as "Summary of 2024 Planning Completion," "2025 Budget Management Plan," "Market Expansion Effectiveness Analysis," and "Internal Audit Issue Rectification." Each chapter is described in natural language, without fixed task division markers, making it a typical unstructured document. After being input into a multi-level knowledge graph engine, such as... Figure 2 As shown, the process of using a semantic parser to transform it into a structured task carrier that can be used for task allocation includes:

[0057] 1. Semantic Parsing and Key Information Extraction: The semantic parser uses natural language processing techniques (such as using pre-trained models like BERT for keyword recognition, entity linking, and chapter structure parsing) to extract three types of key information related to task allocation:

[0058] (1) Document chapter structure information: Identify the hierarchical structure such as “Summary of the completion of the 2024 annual plan → Sub-item 1: Statistics on the achievement rate of planning indicators of each business line; Sub-item 2: Analysis of the reasons for non-compliance indicators; Sub-item 3: Formulation of subsequent improvement measures”, and clarify the hierarchical relationship of task decomposition;

[0059] (2) Exclusive organizational unit identifier: Identify the statement "The Finance Department is exclusively responsible for budget preparation and cost accounting" in the "2025 Budget Management Plan" section and extract the exclusive identifier of "Finance Department"; Identify the statement "The Audit Department takes the lead and various business departments cooperate in rectification" in the "Internal Audit Issue Rectification" section and extract the organizational unit association identifier of "Audit Department (leading) + Business Department (cooperating)";

[0060] (3) Core task description terms: Extract core task terms such as “competitor market share research”, “customer satisfaction statistics” and “new region expansion effect evaluation” from the “Market expansion effectiveness analysis” chapter, and clarify the specific task content corresponding to each chapter.

[0061] The process of transforming the unstructured report into a structured task carrier in this embodiment of the invention is as follows: Based on the extracted key information, the unstructured report is transformed into a "structured task tree," a carrier that can be used for task allocation. The task tree takes the "Overall Task of the 2024 Corporate Work Report" as the root node, with each chapter corresponding to a first-level sub-node (such as "Planning Completion Status Task" and "Budget Management Task"). The sub-items under each chapter correspond to second-level / third-level sub-nodes with the core tasks, and each task node is marked with associated key information (such as "Budget Management Task - First-level Sub-node: Dedicated Organizational Unit = Finance Department; Core Task Terminology = Budget Preparation, Cost Accounting" and "Competitor Market Share Research - Second-level Sub-node: Core Task Terminology = Competitor Data Collection, Market Share Calculation; Associated Organizational Unit = Marketing Department"), forming a hierarchical and complete structured task allocation basis.

[0062] The above process addresses the pain point of "inefficient manual task breakdown," improving task breakdown efficiency and standardization. Semantic parsing can convert unstructured documents into structured task trees within minutes, far faster than the 2-5 days required for manual breakdown. Simultaneously, the parsing process, based on unified keyword recognition and structure parsing rules, avoids subjective differences inherent in manual breakdown, ensuring consistent task breakdown standards across different reports and time periods, forming a standardized task breakdown process suitable for scenarios with a surge in report volume or increased content complexity. It provides precise basis for subsequent multi-level task allocation: key information such as "exclusive organizational unit identifiers" and "core task terms" marked on the structured task tree directly supports the departmental responsibility matching in "graph construction" and the employee capability matching in "federated allocation" in subsequent embodiments of this invention, avoiding allocation deviations due to missing information and improving task allocation accuracy.

[0063] Step S2: Construct a two-layer knowledge graph covering both organizational unit and individual member dimensions. Use cross-dimensional association algorithms to model cross-level relationships, map the organizational unit dimension graph and the individual member dimension graph to a unified semantic space, and associate the two types of graphs.

[0064] Specifically, in this embodiment of the invention, the organizational unit dimension graph uses organizational units as nodes and the relationship between rights and responsibilities between units as edges, while the individual member dimension graph uses individual members as nodes and the collaborative relationship between individuals as edges. The relationship between rights and responsibilities between organizational units includes approval relationships and collaborative relationships, and the collaborative relationship between individual members uses historical collaboration frequency as a quantitative indicator.

[0065] Taking a company's processing of its 2024 Annual Work Report, which involves four organizational units—Finance Department, Planning Department, Marketing Department, and Audit Department—and their respective employees, as an example, this paper elaborates on the construction and relational logic of a two-layer knowledge graph:

[0066] 1. Example of an organizational unit dimension map

[0067] (1) Node definition: The core nodes are "Finance Department, Planning Department, Marketing Department and Audit Department". Each node contains basic attributes such as organizational unit name and core authority and responsibility scope (e.g., Finance Department attributes: Name = Finance Department, core authority and responsibility = budget preparation, cost accounting, fund management; Planning Department attributes: Name = Planning Department, core authority and responsibility = planning formulation, indicator monitoring, progress tracking).

[0068] (2) Definition of Edge (Rights and Responsibilities Association): Taking the approval and collaboration relationships between organizational units as edges, the association types and business meanings are clearly defined:

[0069] Approval relationships: Finance Department → Procurement Department (in this scenario, the Procurement Department participates in the "Procurement Cost Analysis" subtask in the report), with the side attribute labeled "Budget Approval Authority," indicating that the "Procurement Cost Analysis Data" submitted by the Procurement Department must be approved by the Finance Department before being included in the report; Finance Department → Planning Department, with the side attribute labeled "Budget Adjustment Approval Authority," indicating that the "2025 Annual Planning Budget" formulated by the Planning Department must be approved by the Finance Department;

[0070] Collaboration Relationship: Planning Department → Marketing Department, with the side attribute labeled "Planning Collaboration", indicates that the Planning Department needs to collaborate with the Marketing Department to obtain market expansion data in the "Summary of Planning Completion Status" task; Audit Department → Finance Department / Planning Department / Marketing Department, with the side attribute labeled "Audit Rectification Collaboration", indicates that the Audit Department needs the cooperation of each department to provide rectification materials in the "Internal Audit Issue Rectification" task.

[0071] 2. Example of an individual member dimension map

[0072] (1) Node definition: The core employees in each organizational unit are the nodes. Each node includes attributes such as employee name, department, and core competency tags (e.g., Finance Department employee A attributes: name = employee A, department = finance department, core competency = budget preparation (proficiency 0.9), cost accounting (proficiency 0.8); Marketing Department employee B attributes: name = employee B, department = marketing department, core competency = competitor analysis (proficiency 0.9), data statistics (proficiency 0.85)).

[0073] (2) Edge (collaboration relationship) definition: Edges are constructed based on the historical collaboration frequency among employees as a quantitative indicator. The frequency range is 0-1 (the closer the value is to 1, the more frequent the collaboration and the smoother the cooperation).

[0074] Finance Department employee A → Planning Department employee C, with the side attribute labeled "historical collaboration frequency = 0.8", indicates that both were core collaborators in the past 3 "budget and planning collaboration" tasks, and their cooperation efficiency was high;

[0075] Marketing Department Employee B → Marketing Department Employee D, with the side attribute labeled "Historical Collaboration Frequency = 0.9", indicates that the two have worked together for a long time to complete the "Market Data Research and Analysis" task, and have strong data connection and result integration capabilities; Audit Department Employee E → Finance Department Employee A, with the side attribute labeled "Historical Collaboration Frequency = 0.75", indicates that the two have collaborated in 2 "Audit Rectification Data Verification" tasks and are familiar with each other's work processes.

[0076] This invention constructs a two-layer graph that, for the first time, achieves a hierarchical association between "organizational units (departments) - individual members (employees)," unlike existing technologies that only support single-level allocation. The organizational unit graph clarifies the boundaries of departmental responsibilities and collaborative relationships, providing a basis for "reporting tasks → department matching." The individual member graph quantifies employee collaboration efficiency, supporting "departmental tasks → employee matching," forming a complete allocation chain of "report breakdown → department matching → employee matching." This avoids the manual coordination costs caused by broken task allocation chains and improves task execution efficiency. Simultaneously, the organizational unit graph structurally stores the "scope of responsibilities and cross-departmental association rules" scattered across departments, while the individual member graph systematically integrates employee "ability tags and historical collaboration data." These two types of graphs break down departmental data barriers, avoiding information asymmetry caused by data silos and providing a structured data carrier for privacy-preserving sharing under subsequent federated learning mechanisms.

[0077] The cross-dimensional association algorithm used in this embodiment of the invention is the TransR algorithm. The TransR algorithm is used for cross-level relationship modeling, specifically including: for entities h and t in the organizational unit dimensional graph and the cross-level relationship vector r, and entities hr and tr in the member individual dimensional graph and the relationship vector r, the entities are projected from the entity space to the relationship space using the relationship projection matrix Mr. The association relationship satisfies... ; where h r = h M r, t r = t M r Mr ∈ R k×d Let be the projection matrix corresponding to the relation vector r, k be the entity vector dimension, and d be the relation vector dimension.

[0078] This invention employs the TransR algorithm for cross-level relationship modeling between organizational unit dimensional graphs (departments) and individual member dimensional graphs (employees). The core principle is to map entity vectors from two different semantic spaces to a unified relation space using a relation projection matrix, achieving a precise association between "departmental responsibilities and employee capabilities." Taking the cross-level association between "Finance Department (organizational unit)" and "Employee A (individual member)" in a company's 2024 annual work report as an example, the specific application and value of the TransR algorithm are explained. The specific steps are as follows:

[0079] 1. Define the modeling object and vector dimensions

[0080] (1) Entity and Relationship Definition: Organizational Unit Dimension Entity h: Finance Department (Organizational Unit Entity), its original entity vector (Let (k=5) represent the five core responsibilities of "budget preparation, cost accounting, fund management, tax declaration, and financial auditing", and let the vector value be the proficiency of each responsibility, such as (h = [0.95, 0.9, 0.85, 0.8, 0.75]));

[0081] (2) Entity t of the individual member dimension: Employee A (employee of the finance department), whose original entity vector (In the same dimension (k=5), it represents the proficiency of the corresponding rights and responsibilities, such as (t = [0.92, 0.88, 0.8, 0.75, 0.7]));

[0082] Cross-level relationship r: "Budgeting responsibility-capability association" between "Finance Department - Employee A", relationship vector (Let (d=3) represent the three related indicators of "matching of rights and responsibilities, closeness of collaboration, and task completion rate", such as (r = [0.9, 0.85, 0.88])).

[0083] (3) Definition of projection matrix: For the specific relationship r of "budget preparation authority-capability association", a preset relationship projection matrix is ​​defined. (5) 3) A matrix is ​​used to project entities from a 5-dimensional entity space to a 3-dimensional relation space. The matrix values ​​are obtained by training on historical "departmental responsibilities - employee capabilities" correlation data, as shown in the example below:

[0084]

[0085] 2. Perform cross-spatial projection calculations

[0086] According to the TransR algorithm rules, the organizational unit entity h and the individual member entity t are mapped to the relation space through projection matrices, respectively, to obtain the projected entity vectors hr and tr:

[0087] Finance Department Projection Vector h r =h Substituting h = [0.95, 0.9, 0.85, 0.8, 0.75] with Mr above, we calculate hr. [0.91, 0.86, 0.89] (Specific calculation process: see the first column 0.95) 0.85 + 0.9 0.1 +0.85 0.05 + 0.8 0.03 + 0.75 0.02 0.91, and the same applies to the other columns); Employee A's projection vector t r =t Substituting t = [0.92, 0.88, 0.8, 0.75, 0.7] with Mr above, we calculate t. r [0.88, 0.83, 0.86].

[0088] 3. Verify the validity of cross-level associations.

[0089] According to the core correlation formula h of the TransR algorithm r + t r Verify whether the cross-level association between "Finance Department - Employee A" matches:

[0090] Calculate h r+ : [0.91 + 0.9, 0.86 + 0.85, 0.89 + 0.88] = [1.81, 1.71, 1.77];

[0091] Calculate vector similarity (e.g., cosine similarity): h r + With t r The cosine similarity is approximately 0.98 (close to 1), which satisfies the "approximately equal" condition. This indicates that the "budget preparation authority and responsibility of the finance department" and "budget preparation ability of employee A" are correlated and matched, and the "budget management task" can be further assigned from the finance department to employee A.

[0092] This invention uses the TransR algorithm for cross-level relationship modeling to address the pain point of "cross-level semantic misalignment," achieving precise alignment of responsibilities and capabilities. Existing technologies lack cross-level relational modeling. The TransR algorithm maps departments (entity space) and employees (entity space) to a unified relational space through a projection matrix, eliminating semantic dimension differences between the two types of entities. For example, in the above example, the similarity of association between "finance department responsibilities" and "employee A's capabilities" reaches 0.98, ensuring consistency between "responsibilities-capabilities" in task allocation from department to employee, and improving task allocation accuracy.

[0093] The core objective of this invention is to achieve a three-level linkage of "report decomposition → department matching → employee matching". The TransR algorithm is the key technical support: by modeling the cross-level association between "department and employee", the result of "report task → department matching" (such as "budget management task matching finance department") can be directly connected to "department → employee matching" (such as finance department matching employee A), forming a complete allocation chain. This solves the shortcomings of existing technologies that only support single-level allocation and require manual connection, and improves the automation level of the task allocation process.

[0094] Step S3: Based on the privacy protection mechanism and the federated learning mechanism, tasks are assigned to organizational units and individual members.

[0095] like Figure 2 As shown, this embodiment of the invention generates a task list through a federated allocation controller and a differential privacy aggregator, specifically including the following steps:

[0096] S31, calculate the correlation degree between each task and the responsibilities of the organizational unit in the structured task carrier, and filter the matching organizational units based on the correlation degree. Specifically, cosine similarity is used to calculate the correlation degree between the task and the responsibilities of the organizational unit, and a correlation degree threshold is set. When the calculated correlation degree is greater than the threshold, the task is determined to be matched with the corresponding organizational unit.

[0097] S32 involves each organizational unit quantifying the individual capabilities of its members locally and encrypting the quantification results using a privacy protection mechanism. Specifically, this privacy protection mechanism is a differential privacy mechanism, which achieves privacy protection by adding Laplace noise to the quantification results of individual member capabilities, as shown below:

[0098] F=E i +Laplace(0,β)

[0099] Where F is the encrypted individual member ability vector; E i Let E be the original capability vector of individual member i. Let Laplace(0,β) represent random noise with a Laplace distribution of position parameter 0 and scale parameter β, used to modify the original capability vector E. i Perform perturbation encryption.

[0100] S33 aggregates the encrypted individual ability data of members from various organizational units to generate the final task allocation list.

[0101] In one example, the structured task carrier is the "2024 Annual Corporate Work Report Task Tree," where the "2025 Budget Management" task node is labeled with the core terms "budget preparation, cost accounting," and the "Market Expansion Effectiveness Analysis" task node is labeled with the core terms "competitive market share research, customer satisfaction statistics." In the organizational unit dimension diagram, the Finance Department's responsibility vector includes "budget preparation (weight 0.95), cost accounting (weight 0.9)," and the Marketing Department's responsibility vector includes "competitive market share research (weight 0.92), customer satisfaction statistics (weight 0.88)." The process includes the following:

[0102] 1. Calculate the correlation between tasks and organizational unit responsibilities using cosine similarity:

[0103]

[0104] The correlation score between budget management tasks and the finance department is approximately 0.98; the correlation score between budget management tasks and the marketing department is approximately 0; and the correlation score between market expansion analysis tasks and the marketing department is approximately 0.99. With a preset correlation threshold of 0.7, budget management tasks are matched with the finance department, and market expansion analysis tasks are matched with the marketing department; organizational unit allocation is completed directly without manual intervention.

[0105] 2. Each department completes employee capability vector calculation locally (without uploading raw data to the federal controller):

[0106] Finance Department Local: Employee A's Capability Vector E A = [Budgeting: 0.92, Cost Accounting: 0.88], Employee B's Capability Vector EB = [Budgeting: 0.85, Cost Accounting: 0.9]);

[0107] Marketing Department Local: Employee C's Capability Vector E C = [Competitive Product Research: 0.93, Satisfaction Statistics: 0.89], Employee D Capability Vector E D = [Competitive Product Research: 0.87, Satisfaction Statistics: 0.91].

[0108] Differential privacy encryption: Each department adds noise to the capability vector based on the Laplace mechanism. The noise parameter β is set according to the department's privacy level (e.g., β=0.1 for the finance department due to data sensitivity; β=0.08 for the marketing department).

[0109] Finance Department: Employee A's encrypted vector (F) A = E A + Laplace(0,0.1) = [0.92+0.07, 0.88-0.05] = [0.99, 0.83]; Employee B's encrypted vector F B = [0.85-0.06, 0.9+0.04] = [0.79,0.94];

[0110] Marketing Department: Employee C, encrypted vector F C = [0.93+0.06, 0.89-0.03] = [0.99, 0.86]; Employee D's encrypted vector F D = [0.87-0.04, 0.91+0.05] = [0.83, 0.96].

[0111] 3. Aggregate encrypted data and generate the final task assignment list:

[0112] Encrypted data upload and aggregation: The finance department and marketing department will only upload the encrypted data. A F B F C F D Uploaded to the Federal Assignment Controller, the controller does not need to obtain the original capability data, but directly calculates the "task-employee" matching degree based on the encrypted vector (matching degree = cosine similarity between the core terms of the task and the encrypted capability vector):

[0113] Budget management task - Employee A: Match rate ≈ 0.97; Budget management task - Employee B: Match rate ≈ 0.92;

[0114] Market expansion analysis task - employee C: matching degree ≈ 0.98; Market expansion analysis task - employee D: matching degree ≈ 0.93.

[0115] Generate a task list: Sort tasks in descending order of matching degree, filter the top-performing employees to assign tasks, and finally divide the tasks into results: "Budget Management Task → Finance Department Employee A; Market Expansion Effectiveness Analysis Task → Marketing Department Employee C", and synchronize them to the corresponding employee's task terminal. Furthermore, to protect task and employee information, generate a task list including task and employee IDs, for example: {Task ID: T001, T001, Assigned Employee: {Employee ID: E023, Matching Degree: 0.97}, {Employee ID: E107, Matching Degree: 0.93}}.

[0116] This invention utilizes a federated learning framework, enabling departments to upload only encrypted data while retaining the original capability data locally. This breaks down data silos (achieving cross-departmental data collaboration) and ensures privacy through Laplace noise (e.g., a finance department employee's original capability of 0.92 is only visible to the department, while external entities only obtain the encrypted version of 0.99). The core advantage of the differential privacy mechanism is that encryption does not affect data availability. By adding Laplace noise in this invention, the deviation between the encrypted and original vectors is controlled within 5% (e.g., employee A's original budget preparation capability is 0.92, and the encrypted version is 0.99, resulting in a deviation of approximately 7.6%, which can be further reduced by decreasing β). This ensures that the matching degree calculated by the federated controller based on encrypted data still reflects the employee's true capability, avoiding allocation failure due to excessive encryption and achieving a balance between privacy protection and allocation efficiency.

[0117] Step S4: Obtain feedback data on individual member tasks and dynamically adjust the task allocation weights of individual members to achieve overall system load balancing.

[0118] This invention achieves overall system load balancing through a dynamic adjuster. Specifically, it includes the following steps:

[0119] S41, Obtain feedback data on individual member task execution, including task completion progress, task completion quality, and task processing time, to update the individual member's ability quantification results and task backlog.

[0120] S42, dynamically adjusts its task allocation weights based on the backlog amount, using the following formula for weight updates:

[0121]

[0122] Where n represents the total number of members in the current system who can be assigned tasks; Q i The number of tasks to be processed for member individual i; Q represents the total number of tasks to be processed by all individual members. avgThe average number of tasks to be processed for all individual members; η represents the learning rate, which is dynamically adjusted based on system performance.

[0123] η t =η t-1 exp(-0.1 |Q avg - Q i |);

[0124] Where, η t η represents the learning rate at the current moment, a key parameter used to adjust the task allocation weights for employees at time t. The learning rate determines the step size of the weight adjustment, and its value affects the magnitude of the task allocation weight adjustment; t-1 This represents the learning rate at the previous time step, specifically the value at time step t-1. It provides a basic reference value for calculating the learning rate at the current time step, reflecting the continuity and correlation of the learning rate over time. avg Q is the average number of tasks to be processed by all individual members. It is obtained by dividing the sum of the number of tasks to be processed by all individual members with assignable tasks in the system by the total number of individual members. It is used to measure the average task load level of the entire system. i This represents the number of tasks pending for individual member i, reflecting the number of unfinished tasks currently backlogged for a single employee. It is determined by comparing the Q values ​​of different employees. i With Q avg This allows us to determine the employee's workload, whether it is overloaded, average, or underloaded; exp() is an exponential function with the natural constant e as its base, which converts the difference in employee task backlog into a learning rate adjustment coefficient. In this embodiment of the invention, η is set with preset upper and lower limits to prevent oscillation.

[0125] This invention uses the task load adjustment of "3 employees (employees C, D, and E) in the marketing department" when a company processes the "2024 Annual Company Work Report" as an example to illustrate the application and value of the weight update formula.

[0126] 1. Basic Parameter Definitions

[0127] The total number of individual members is n=3 (Marketing Department employees C, D, and E all participate in the allocation of sub-tasks related to "Market Expansion Effectiveness Analysis").

[0128] Number of tasks to be processed: Q of employee C C =8. Employee D's Q D =3. Q of employee E E =1;

[0129] Initial learning rate η t-1 =0.5 (preset upper limit η)max =0.8, lower limit η min =0.2).

[0130] 2. Calculation of key parameters

[0131] Total number of tasks to be processed by all members = 8 + 3 + 1 = 12;

[0132] Average number of tasks to be processed = 4.

[0133] 3. Learning rate calculation (based on η) t =η t-1 xp(-0.1 |Q avg - Q i |))

[0134] Employee C (Q) C =8, Overload): η t =0.5 exp(-0.1 ≈0.335, within the range of (0.2-0.8), is valid.

[0135] Employee D (Q) D =3, load is close to average): η t =0.5 exp(-0.1 ≈0.45;

[0136] Employee E (Q E =1, load too low): η t =0.5 exp(-0.1 ≈0.37.

[0137] 4. Adjustment of task allocation weights

[0138] Assume that the weights of the three people before the adjustment were all w. t-1 =0.3, the weight update logic is "the higher the load, the more the weight decreases" (combined with η). t (Control the adjustment range)

[0139] Employee C (high load, η) t ≈0.335)):w t = 0.33 - 0.335 0.2 ≈ 0.26 (Weight reduction, fewer new tasks assigned);

[0140] Employee D (moderate load, η) t ≈0.45): wt = 0.33 - 0.45 0.05 ≈ 0.31) (weight fine-tuning);

[0141] Employee E (low load, η) t ≈0.37): w t =0.33 + 0.37 0.2 ≈ 0.39 (weight increase, additional new task allocation).

[0142] This invention uses a formula to dynamically calculate the learning rate and weights. Employees with excessive workloads (such as C) have their weights reduced, while employees with insufficient workloads (such as E) have their weights increased. This ensures that tasks are tilted towards idle members, avoids uneven workloads, reduces waste of human resources, improves overall task processing efficiency, and achieves the technical effect of a response time of less than 30 seconds when the system experiences a sudden increase in task volume of 200%. This also avoids adjustment fluctuations and ensures stable system operation.

[0143] This embodiment also provides a task decomposition and federated allocation system based on a multi-level knowledge graph. This system is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0144] This embodiment provides a task decomposition and federated allocation system based on a multi-level knowledge graph, such as... Figure 3 As shown, it includes:

[0145] The document structuring module 31 is used to use semantic parsing to parse unstructured documents that need to be completed through cross-departmental collaboration, extract key information related to task allocation from the documents, and transform the unstructured documents into structured task carriers that can be used for task allocation.

[0146] The two-layer knowledge graph construction module 32 is used to construct a two-layer knowledge graph covering the organizational unit dimension and the individual member dimension. It uses a cross-dimensional association algorithm to model cross-level relationships, maps the organizational unit dimension graph and the individual member dimension graph to a unified semantic space, and associates the two types of graphs.

[0147] Task allocation module 33 is used to allocate tasks to organizational units and individual members based on a privacy protection mechanism combined with a federated learning mechanism.

[0148] The dynamic adjustment module 34 is used to obtain feedback data on the tasks performed by individual members and dynamically adjust the task allocation weights of individual members in order to achieve overall system load balancing.

[0149] In some optional implementations, the key information related to task allocation in the document structuring module 31 includes at least one of the following: document chapter structure, exclusive organizational unit identifier, and core task description terminology; the structured task carrier is a structured task tree, and when constructing the structured task tree, key fields and task nodes are extracted from the unstructured document to be processed, and the task tree structure is built based on the association between the key fields and task nodes.

[0150] In some optional implementations, the organizational unit dimension graph in the two-layer knowledge graph construction module 32 uses organizational units as nodes and the rights and responsibilities between units as edges, while the individual member dimension graph uses individual members as nodes and the collaborative relationships between individuals as edges; the rights and responsibilities between organizational units include approval relationships and collaborative relationships, and the collaborative relationships between individual members use historical collaboration frequency as a quantitative indicator.

[0151] The cross-dimensional association algorithm is the TransR algorithm. It is used to model cross-level relationships, specifically including: for entities h and t in the organizational unit dimension graph and the cross-level relationship vector r, and entities hr and tr in the member individual dimension graph and the relationship vector r, the entities are projected from the entity space to the relationship space using the relationship projection matrix Mr. The association relationship satisfies h r + ;t r；

[0152] Among them, h r = h M r, t r = t M r Mr ∈ R k×d Let be the projection matrix corresponding to the relation vector r, k be the entity vector dimension, and d be the relation vector dimension.

[0153] In some optional implementations, the task allocation module 33 includes:

[0154] The organization unit matching submodule is used to calculate the correlation between each task in the structured task carrier and the rights and responsibilities of the organization unit, and to filter the matching organization units based on the correlation. The encryption processing submodule is used to quantify the individual capabilities of members locally by each organization unit and to encrypt the quantification results through a privacy protection mechanism. The assignment list generation submodule is used to aggregate the encrypted individual capability data of members from each organization unit and generate the final task assignment list.

[0155] In one optional implementation, cosine similarity is used to calculate the correlation between the task and the responsibilities of the organizational unit, and a correlation threshold is set. When the calculated correlation is greater than the threshold, the task is determined to match the corresponding organizational unit.

[0156] For tasks with a unique organizational unit identifier, they are directly associated with the corresponding organizational unit.

[0157] The privacy protection mechanism is a differential privacy mechanism, which achieves privacy protection by adding Laplace noise to the quantification results of individual member abilities, as shown below:

[0158] F=E i +Laplace(0,β)

[0159] Where F is the encrypted individual member ability vector; E i Let E be the original capability vector of individual member i. Let Laplace(0,β) represent random noise with a Laplace distribution of position parameter 0 and scale parameter β, used to modify the original capability vector E. i Perform perturbation encryption.

[0160] In one alternative implementation, the dynamic adjustment module 34 includes:

[0161] The feedback data submodule is used to obtain feedback data on individual member tasks, including task completion progress, task completion quality, and task processing time, and is used to update the individual member's ability quantification results and task backlog.

[0162] The weight update submodule is used to dynamically adjust the task allocation weights based on the backlog, using the following formula for weight updates:

[0163]

[0164] Where n represents the total number of members in the current system who can be assigned tasks; Q i The number of tasks to be processed for member individual i; Q represents the total number of tasks to be processed by all individual members. avg The average number of tasks to be processed for all individual members; η represents the learning rate, which is dynamically adjusted based on system performance.

[0165] η t =η t-1 exp(-0.1 |Q avg - Q i |);

[0166] η t η represents the learning rate at the current moment.t-1 This represents the learning rate at the previous time step, and η has preset upper and lower limits.

[0167] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0168] In this embodiment, the task decomposition and federation system based on a multi-level knowledge graph is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit), a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0169] This invention also provides a computer device having the above-described features. Figure 3 The system shown is a task decomposition and federated allocation system based on a multi-level knowledge graph.

[0170] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.

[0171] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0172] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0173] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0174] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0175] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0176] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0177] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0178] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for task decomposition and federated allocation based on a multi-level knowledge graph, characterized in that, include: Semantic parsing is used to parse unstructured documents that require cross-departmental collaboration, extract key information related to task allocation from the documents, and transform the unstructured documents into structured task carriers that can be used for task allocation. A two-layer knowledge graph covering both organizational unit and individual member dimensions is constructed. Cross-dimensional association algorithms are used to model cross-level relationships, mapping the organizational unit dimension graph and the individual member dimension graph to a unified semantic space for association. The organizational unit dimension graph uses organizational units as nodes and the rights and responsibilities relationships between units as edges, while the individual member dimension graph uses individual members as nodes and the collaboration relationships between individuals as edges. The rights and responsibilities relationships between organizational units include approval relationships and collaboration relationships, and the collaboration relationships between individual members are quantified using historical collaboration frequency. The cross-dimensional association algorithm is the TransR algorithm. It is used to model cross-level relationships, specifically including: for entities h and t in the organizational unit dimension graph and the cross-level relationship vector r, and entities hr and tr in the member individual dimension graph and the relationship vector r, the entities are projected from the entity space to the relationship space using the relationship projection matrix Mr. The association relationship satisfies h r + t r ; where h r = h M r, t r = t M r Mr ∈ R k×d Let be the projection matrix corresponding to the relation vector r, k be the entity vector dimension, and d be the relation vector dimension; Based on a privacy protection mechanism combined with a federated learning mechanism, task allocation is performed for organizational units and individual members. Cosine similarity is used to calculate the correlation between the task and the responsibilities of the organizational unit, and a correlation threshold is set. When the calculated correlation is greater than the threshold, the task is determined to be matched with the corresponding organizational unit. For tasks with a unique organizational unit identifier, they are directly associated with the corresponding organizational unit. The privacy protection mechanism is a differential privacy mechanism, which achieves privacy protection by adding Laplace noise to the quantification results of individual member abilities, as follows: F=E i +Laplace(0,β) Where F is the encrypted individual member ability vector; E i Let E be the original capability vector of individual member i. Let Laplace(0,β) represent random noise with a Laplace distribution of position parameter 0 and scale parameter β, used to modify the original capability vector E. i Perform perturbation encryption; Obtain feedback data on individual member tasks and dynamically adjust the task allocation weights for each member to achieve overall system load balancing, including: Obtain feedback data on individual member task execution, including task completion progress, task completion quality, and task processing time, to update the individual member's ability quantification results and task backlog. The task allocation weights are dynamically adjusted based on the backlog, and the weights are updated using the following formula: Where n represents the total number of members in the current system who can be assigned tasks; Q i The number of tasks to be processed for member individual i; Q represents the total number of tasks to be processed by all individual members. avg The average number of tasks to be processed for all individual members; η represents the learning rate, which is dynamically adjusted based on system performance. the t =η t-1 exp(-0.1 |Q avg - Q i |); Where, η t η represents the learning rate at the current moment. t-1 η represents the learning rate at the previous time step, and η has preset upper and lower limits. exp() is an exponential function with the natural constant e as its base.

2. The method according to claim 1, characterized in that, The key information related to task allocation includes at least one of the following: document chapter structure, exclusive organizational unit identifier, and core task description terminology; the structured task carrier is a structured task tree. When constructing the structured task tree, key fields and task nodes are extracted from the unstructured document to be processed, and the task tree structure is built based on the relationship between the key fields and task nodes.

3. The method according to claim 1, characterized in that, The method, which combines privacy protection mechanisms with federated learning, utilizes key information related to task allocation to allocate tasks to organizational units and individual members, including: Calculate the correlation between each task and the responsibilities of the organizational unit in the structured task carrier, and filter the matching organizational units based on the correlation; each organizational unit quantifies the individual capabilities of its members locally, and encrypts the quantification results through a privacy protection mechanism; aggregate the encrypted individual capability data of members from each organizational unit to generate the final task allocation list.

4. A task decomposition and federated allocation system based on a multi-level knowledge graph, wherein the method described in claim 1 is characterized in that, include: The document structuring module is used to parse unstructured documents that require cross-departmental collaboration using semantic parsing methods, extract key information related to task allocation from the documents, and transform the unstructured documents into structured task carriers that can be used for task allocation. The two-layer knowledge graph construction module is used to construct a two-layer knowledge graph covering the organizational unit dimension and the individual member dimension. It uses cross-dimensional association algorithms to model cross-level relationships, maps the organizational unit dimension graph and the individual member dimension graph to a unified semantic space, and associates the two types of graphs. The task allocation module is used to allocate tasks to organizational units and individual members based on a privacy protection mechanism combined with a federated learning mechanism. The dynamic adjustment module is used to obtain feedback data on the tasks performed by individual members and dynamically adjust the task allocation weights of individual members in order to achieve overall system load balancing.

5. A computer device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the task decomposition and federated allocation method based on a multi-level knowledge graph as described in any one of claims 1 to 3 by executing the computer instructions.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the task decomposition and federated allocation method based on a multi-level knowledge graph as described in any one of claims 1 to 3.

7. A computer program product, characterized in that, It includes computer instructions for causing a computer to execute the task decomposition and federated allocation method based on a multi-level knowledge graph, as described in any one of claims 1 to 3.

Images