A method and apparatus for chapter task aggregation based on semantic entanglement

Abstract

This invention relates to the field of computer technology and discloses a method and apparatus for chapter task aggregation based on semantic entanglement. The method includes: generating a report graph based on a target report; determining the business type of nodes in the report graph and dividing them into similar task groups; calculating the quantum entanglement degree between any two nodes in the similar task group to obtain the semantic entanglement degree of the nodes; calculating the merging score of nodes when the semantic entanglement degree is greater than a preset threshold to determine whether nodes should be merged or split; performing conflict detection during splitting to obtain the conflict type; splitting the task into split task groups according to the splitting method corresponding to the conflict type; generating sub-tasks of the split task groups from the total tasks of the similar task groups; and aggregating the report chapters corresponding to the nodes into the sub-tasks. This invention achieves initial aggregation through the division of similar task groups and further aggregation using semantic entanglement, reducing the number of tasks and accurately aggregating report chapters to the corresponding tasks, thus achieving refined aggregation.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and specifically to a method and apparatus for chapter task aggregation based on semantic entanglement. Background Technology

[0002] In enterprise reporting systems, a single comprehensive report may contain anywhere from a dozen to hundreds of independent chapters. Traditional systems generate a separate task for each chapter, leading to a task overload on the receiving end. When a single report has more than 20 chapters, the task generation by traditional systems can cause a 300% surge in user clicks, significantly increasing the user's workload and time costs.

[0003] To address the aforementioned issues, existing chapter aggregation technology supports aggregation from a time dimension, but this only enables simple task aggregation, resulting in a large number of generated to-do tasks and inaccurate categorization. Summary of the Invention

[0004] In view of this, the present invention provides a chapter task aggregation method and apparatus based on semantic entanglement to solve the problem that existing chapter aggregation methods only aggregate from the time dimension, resulting in a large number of to-do tasks and inaccurate classification.

[0005] In a first aspect, the present invention provides a chapter task aggregation method based on semantic entanglement, the method comprising:

[0006] A report graph is generated based on the target report. The report graph includes multiple nodes and their characteristics. Nodes represent report sections in the target report.

[0007] The business type of each node in the report graph is determined, and based on the business types of all nodes, multiple similar to-do groups are obtained;

[0008] For each group of tasks of the same type, based on the node characteristics of each node, the quantum entanglement degree between any two nodes in the group of tasks of the same type is calculated to obtain the semantic entanglement degree of each node.

[0009] When the semantic entanglement of a node is greater than a preset threshold, the node’s merging score is calculated, and based on the merging score, it is determined whether the node should be merged or split.

[0010] When splitting a node, conflict detection is performed on the node and similar to-do groups to obtain the conflict type;

[0011] Based on the splitting method corresponding to the conflict type, the same type of to-do group is split into multiple split to-do groups. In the total to-do tasks corresponding to the same type of to-do group, the sub-to-do tasks corresponding to each split to-do group are generated respectively. The report chapters corresponding to the nodes are aggregated into the sub-to-do tasks of the corresponding split to-do group.

[0012] This invention transforms the target report into a report graph, explicitly displaying the chapter information and relationships between chapters. It determines the business type of each node in the graph and categorizes them into similar task groups, achieving initial chapter aggregation. Targeted analysis is then performed on each similar task group. For each node within the group, the semantic entanglement degree is calculated based on its correlation with other nodes in the group. Based on this initial aggregation, nodes with high correlation are quickly selected according to the semantic entanglement degree and a preset threshold, achieving coarse screening of node associations. A merging score is calculated for the coarsely screened nodes, and the merging score determines whether to merge or split the nodes, enabling refined decision-making. Conflict detection is performed during node splitting, and conflict types are determined. Sub-tasks are generated based on conflict types, reducing the number of tasks while ensuring compliance. Initial aggregation is achieved through similar task group division, and further aggregation is performed using semantic entanglement, reducing the number of generated tasks and accurately aggregating each report chapter into appropriate tasks, achieving refined aggregation and effectively reducing the user's operational burden and time cost.

[0013] In one alternative implementation, after determining whether nodes are merged or split, the method further includes:

[0014] When merging nodes, determine whether the node has compatible permissions with similar to-do groups;

[0015] When a node's permissions are compatible with those of a similar to-do group, the report section corresponding to the node will be aggregated into the total to-do task corresponding to the same to-do group.

[0016] When a node's permissions are incompatible with any node in the same type of to-do group, the same type of to-do group is split into multiple split to-do groups according to permissions. Sub-to-do tasks corresponding to each split to-do group are generated in the total to-do tasks corresponding to the same type of to-do group. The report chapters corresponding to the node are aggregated into the sub-to-do tasks of the corresponding split to-do group.

[0017] This embodiment mitigates permission conflicts by assessing permission compatibility with similar to-do groups during node merging, ensuring the merging operation complies with enterprise permission standards. When permissions are compatible, nodes are aggregated into the total to-do tasks of the similar to-do group, streamlining the number of to-dos and improving user processing efficiency. When permissions are incompatible, nodes are split according to permissions and sub-to-do tasks are generated to balance permission security and business integrity, avoiding permission conflicts.

[0018] In one optional implementation, generating a report map based on the target report includes:

[0019] Each report section in the target report is treated as a node, and the section features of each report section are extracted and used as the node features of the corresponding node.

[0020] When there is a reference relationship between any two nodes, and / or the semantic similarity between two nodes is greater than a preset threshold, and / or there is a business logic relationship between two nodes, an edge is constructed between the two nodes;

[0021] A graph structure is constructed based on the nodes corresponding to each chapter of the target report, the node characteristics of each node, and the edges between the nodes.

[0022] Multiple sample report chapters are obtained, and a sample graph structure is constructed based on the multiple sample report chapters. The pre-trained graph neural network is fine-tuned based on the sample graph structure to obtain the target graph neural network. Each sample report chapter carries sample annotations.

[0023] A target graph neural network is used to update the node features of each node in the graph structure based on the node features of the node's neighboring nodes.

[0024] The report graph is obtained based on each updated node in the graph structure.

[0025] This embodiment constructs a graph structure corresponding to the target report to make the information of the report chapters and the relationships between them explicit. It then uses a fine-tuned target graph neural network to update the node features of each node in the graph structure. This ensures that the node features of each node not only retain their own business attributes but also incorporate the contextual relationships of neighboring nodes, further enhancing the ability of node features to represent the business relationships between chapters.

[0026] In one alternative implementation, determining the service type of each node in the report graph includes:

[0027] A natural language processing classifier is used to determine the initial type of a node based on the node features of each node in the report graph.

[0028] In the business rule engine, the rule corresponding to the initial type is called to validate the keywords and chapter content in the node characteristics of the node respectively;

[0029] Once the keywords and chapter content of a node have passed validation, the initial type will be determined as the business type of the node.

[0030] If the keyword or chapter content of a node fails the verification, the initial type is manually reviewed to obtain the business type of the node.

[0031] This embodiment uses a natural language processing classifier to initially predict the initial type of a node, and then calls the corresponding rules for verification. This dual verification improves the accuracy of business type classification.

[0032] In one optional implementation, for each group of tasks of the same type, based on the node characteristics of each node, the quantum entanglement degree between any two nodes in the group of tasks of the same type is calculated to obtain the semantic entanglement degree of each node, including:

[0033] For any two nodes in each to-do group of the same type, calculate the semantic similarity, workflow similarity, time comparison value and permission conflict value between the two nodes based on the node characteristics of the two nodes;

[0034] The quantum entanglement degree between two nodes is obtained by weighting the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes and their corresponding weights.

[0035] For each node, the maximum value among all quantum entanglement degrees corresponding to the node is taken as the semantic entanglement degree of the node.

[0036] This embodiment quantifies the correlation strength between chapters by calculating the quantum entanglement degree between nodes, and takes the maximum value of the quantum entanglement degree corresponding to the node as the semantic entanglement degree, providing a basis for subsequent coarse screening.

[0037] In one optional implementation, for any two nodes in each similar to-do group, based on the node characteristics of the two nodes, the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes are calculated, including:

[0038] For any two nodes in each to-do group of the same type, determine the semantic similarity between the two nodes based on the chapter title and keywords in the node features of the two nodes;

[0039] Based on the approval paths in the node features of two nodes, identify the common approval nodes and all approval nodes in the approval paths of the two nodes, and determine the ratio of the common approval nodes to all approval nodes as the workflow similarity between the two nodes.

[0040] Based on the deadlines in the node features of two nodes, calculate the difference between their deadlines. Based on the difference between their deadlines and a preset time interval, calculate the time comparison value between the two nodes.

[0041] Based on the approval path in the node characteristics of the two nodes, the permission conflict value between the two nodes is determined.

[0042] This embodiment ensures that quantum entanglement accurately reflects the business relationships between nodes by comprehensively covering the features of report chapters in semantic, process, time, and permission-related dimensions.

[0043] In one alternative implementation, the merging score of computing nodes includes:

[0044] When a node is an isolated node in the report graph, set the value assigned to the node in the business continuity dimension to the first preset value; when a node is not an isolated node in the report graph, set the value assigned to the node in the business continuity dimension to the second preset value.

[0045] When the node's node characteristics do not include a timeliness identifier and the node's business type is not timeliness sensitive, the node's time sensitivity dimension is set to the first preset value. When the node's node characteristics include a timeliness identifier or the node's business type is timeliness sensitive, the node's time sensitivity dimension is set to the second preset value.

[0046] When a node's node characteristics include a single approval node, the node's permission complexity dimension is set to the first preset value; when a node's node characteristics include multiple approval nodes, the node's permission complexity dimension is set to the second preset value.

[0047] Based on the values ​​assigned to nodes in the dimensions of business continuity, time sensitivity, and permission complexity, a multi-dimensional merging matrix corresponding to the nodes is obtained;

[0048] The merged score of a node is obtained by weighting and summing the values ​​assigned to each dimension in the multi-dimensional merge matrix and the preset dimension weights.

[0049] This embodiment constructs a multi-dimensional merging matrix from three dimensions: business continuity, time sensitivity, and permission complexity. Based on this matrix, a merging score is determined, quantifying the merging tendency of individual chapters and providing a basis for judgment on refined chapter aggregation.

[0050] In one optional implementation, determining whether nodes should be merged or split based on the merge score includes:

[0051] Obtain the preset merging probability threshold;

[0052] The merging probability of a node is calculated based on the node's merging score and a preset merging probability threshold.

[0053] Based on the node merging probability, calculate the merging tendency coefficient and the splitting tendency coefficient;

[0054] A random value is generated within a preset range. When the random value is less than the merging tendency coefficient or the merging tendency coefficient is greater than the splitting tendency coefficient, the node is determined to be merged.

[0055] If the random value is greater than the merging tendency coefficient or the merging tendency coefficient is not greater than the splitting tendency coefficient, then the node is determined to be split.

[0056] This embodiment obtains a preset merging probability threshold and calculates the merging probability based on the merging score and the threshold, transforming the discrete merging score into a continuous probability value, which more accurately reflects the possibility of node merging. Then, it calculates the merging tendency coefficient and the splitting tendency coefficient to quantify the tendency of nodes to transform into merging and splitting states. Based on this, it determines whether a node should be split or merged in two ways, which helps to achieve refined aggregation.

[0057] In one alternative implementation, the method further includes:

[0058] Once a target report is detected to have been published, delete all pending tasks corresponding to that target report.

[0059] This embodiment automatically deletes the associated pending tasks of the target report once it is detected that the target report has been published, thus avoiding redundant pending tasks of published reports from consuming system resources or interfering with user operations.

[0060] Secondly, the present invention provides a chapter task aggregation device based on semantic entanglement, the device comprising:

[0061] The generation module is used to generate a report graph based on the target report. The report graph includes multiple nodes and their node features. The nodes represent report sections in the target report.

[0062] The determination module is used to determine the business type of each node in the report graph, and based on the business types of all nodes, divide them into multiple similar to-do groups;

[0063] The computation module is used to calculate the quantum entanglement degree between any two nodes in each to-do group of the same type, based on the node characteristics of each node, and to obtain the semantic entanglement degree of each node.

[0064] The first judgment module is used to calculate the merging score of a node when the semantic entanglement degree of the node is greater than a preset threshold, and to determine whether the node should be merged or split based on the merging score.

[0065] The detection module is used to perform conflict detection on nodes and similar task groups when nodes are split, and to obtain the conflict type.

[0066] The first aggregation module is used to split similar to-do groups into multiple split to-do groups according to the splitting method corresponding to the conflict type, generate sub-to-do tasks corresponding to each split to-do group in the total to-do tasks corresponding to the similar to-do groups, and aggregate the report chapters corresponding to the nodes into the sub-to-do tasks of the corresponding split to-do groups.

[0067] 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 chapter task aggregation method based on semantic entanglement described in the first aspect or any corresponding embodiment thereof.

[0068] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the semantically entangled chapter task aggregation method of the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0069] 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.

[0070] Figure 1 This is a flowchart illustrating the chapter task aggregation method based on semantic entanglement according to an embodiment of the present invention;

[0071] Figure 2 This is a structural block diagram of a chapter task aggregation device based on semantic entanglement according to an embodiment of the present invention;

[0072] Figure 3 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0073] 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.

[0074] Existing chapter aggregation technologies support aggregation based on time, but this only enables simple task aggregation, resulting in a large number of generated to-do tasks with inaccurate categorization. This invention achieves initial aggregation by grouping to-do tasks of the same type, and then uses semantic entanglement for further aggregation, reducing the number of generated to-do tasks. Furthermore, it accurately aggregates each report chapter into appropriate to-do tasks, achieving refined aggregation and effectively reducing the user's operational burden and time costs.

[0075] According to an embodiment of the present invention, a chapter task aggregation method based on semantic entanglement is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system 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 executed in a different order than that shown here.

[0076] This embodiment provides a chapter task aggregation method based on semantic entanglement, which can be used in a reporting system. Figure 1 This is a flowchart of a chapter task aggregation method based on semantic entanglement according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0077] Step S101: Generate a report graph based on the target report. The report graph includes multiple nodes and their node features. Nodes represent report sections in the target report.

[0078] Specifically, the target report includes multiple independent report chapters. The unstructured report chapters are transformed into structured graph data, making the references, semantics, business relationships and other information between the report chapters explicit, so as to support the subsequent chapter aggregation and avoid aggregation deviations caused by isolated chapter information.

[0079] Step S102: Determine the business type of each node in the report graph, and based on the business types of all nodes, divide them into multiple similar to-do groups.

[0080] Specifically, business type is the core basis for ensuring that chapter aggregation aligns with business logic. Clearly define the business type of each report chapter, such as financial auditing, legal compliance, etc., and group report chapters with the same business type together to obtain multiple similar to-do groups, thus achieving initial aggregation of report chapters.

[0081] Step S103: For each group of tasks of the same type, based on the node characteristics of each node, calculate the quantum entanglement degree between any two nodes in the group of tasks of the same type to obtain the semantic entanglement degree of each node.

[0082] Specifically, the quantum entanglement degree between any two nodes in the group is calculated to quantify the correlation strength between nodes and obtain the semantic entanglement degree of the nodes. This helps to quickly screen out nodes with high correlation based on the initial aggregation, thus achieving a coarse screening of node correlation.

[0083] Step S104: When the semantic entanglement degree of a node is greater than a preset threshold, calculate the node's merging score, and based on the merging score, determine whether the node should be merged or split.

[0084] Specifically, if the semantic entanglement of a node is greater than a preset threshold, it indicates that the report chapter corresponding to that node is highly correlated with other report chapters within the group, and has value for further aggregation and evaluation. At this time, the merge score of the report chapter is calculated to make a refined merge decision for that report chapter.

[0085] In some optional implementations, if the semantic entanglement of a node is not greater than a preset threshold, it indicates that the chapter is weakly related to other chapters in the group and no further aggregation evaluation is needed. Since similar to-do groups correspond to a total to-do task in the reporting system (such as "Financial Audit To-Do"), the weakly related chapter (such as "xxx Budget Report") can be made an independent sub-to-do task in the total to-do task. This avoids interfering with the aggregation of strongly related chapters and ensures that the chapter's to-do is not missed, balancing aggregation efficiency and to-do completeness.

[0086] Step S105: When splitting a node, perform conflict detection on the node and similar task groups to obtain the conflict type.

[0087] Specifically, if a report section corresponding to a node needs to be split, a comprehensive conflict detection must first be performed on that node and every node in its corresponding to-do group (excluding nodes with semantic entanglement not exceeding a preset threshold). This includes permission conflict detection, business logic detection, and user-forced split detection. For permission conflict detection, if node A requires approval from the finance department and node B requires approval from the legal department, then there is a permission conflict, and the conflict type is permission conflict. For business logic detection, if node A is a budget section and node B is a legal clause section, then they belong to business logic that cannot be aggregated, and the conflict type is business logic conflict. For user-forced split detection, if a user manually intervenes to create a sub-to-do task for a node, then the conflict type is forced conflict. By identifying the conflict types that lead to the need for section splitting through detection, a clear direction is provided for selecting a reasonable splitting method, avoiding blind splitting.

[0088] Step S106: According to the splitting method corresponding to the conflict type, split the same type of to-do group into multiple split to-do groups, generate the sub-to-do tasks corresponding to each split to-do group in the total to-do tasks corresponding to the same type of to-do group, and aggregate the report chapters corresponding to the nodes into the sub-to-do tasks of the corresponding split to-do group.

[0089] Specifically, based on the conflict type, a corresponding splitting method is matched to divide similar to-do groups into multiple independent split to-do groups. When the conflict type is a permission conflict, similar to-do groups are split according to different permissions, so that the nodes in each split to-do group have the same permissions, such as Finance Department Approval (3 sub-items) and Legal Department Approval (2 sub-items). When the conflict type is a business logic conflict, similar to-do groups are split according to different business logic, so that the nodes in each split to-do group have the same business logic, such as Budget Review (4 sub-items) and Contract Approval (1 sub-item). When the conflict type is a forced conflict, the chapters that are forcibly split are separated into independent split to-do groups, such as Strategic Planning - Final Review. Under the overall to-do task corresponding to similar to-do groups, a dedicated sub-to-do task is generated for each split to-do group, and the report chapters corresponding to each node after splitting are accurately aggregated into the corresponding sub-to-do tasks to form the sub-items of the overall to-do task, realizing refined aggregation management of report chapters.

[0090] This invention transforms the target report into a report graph, explicitly displaying the chapter information and relationships between chapters. It determines the business type of each node in the graph and categorizes them into similar task groups, achieving initial chapter aggregation. Targeted analysis is then performed on each similar task group. For each node within the group, the semantic entanglement degree is calculated based on its correlation with other nodes in the group. Based on this initial aggregation, nodes with high correlation are quickly selected according to the semantic entanglement degree and a preset threshold, achieving coarse screening of node associations. A merging score is calculated for the coarsely screened nodes, and the merging score determines whether to merge or split the nodes, enabling refined decision-making. Conflict detection is performed during node splitting, and conflict types are determined. Sub-tasks are generated based on conflict types, reducing the number of tasks while ensuring compliance. Initial aggregation is achieved through similar task group division, and further aggregation is performed using semantic entanglement, reducing the number of generated tasks and accurately aggregating each report chapter into appropriate tasks, achieving refined aggregation and effectively reducing the user's operational burden and time cost.

[0091] This embodiment provides a chapter task aggregation method based on semantic entanglement, which specifically includes the following steps:

[0092] Step S201: Generate a report graph based on the target report. The report graph includes multiple nodes and their node features. Nodes represent report sections in the target report.

[0093] Specifically, step S201 includes:

[0094] Step S2011: Take each report chapter in the target report as a node, extract the chapter features of each report chapter, and use them as the node features of the corresponding node of the report chapter.

[0095] Specifically, each report section in the target report is treated as an independent node. Section features such as the section title, content, keywords, approval path, approval nodes, and deadline are extracted and used as node features for the corresponding node, thus comprehensively representing the information contained in that section. Optionally, the above section features are merely examples; different content can be extracted as needed. The extraction method is based on existing technology and will not be elaborated further here.

[0096] Step S2012: When there is a reference relationship between any two nodes, and / or the semantic similarity between two nodes is greater than a preset threshold, and / or there is a business logic relationship between two nodes, an edge is constructed between the two nodes.

[0097] Specifically, three conditions are considered when constructing edges between nodes: (1) there is a reference relationship between the report chapters corresponding to the two nodes, such as the chapter content of chapter A indicating that the description in chapter B is provided; (2) the semantic similarity of the chapter titles, chapter content, and keywords of the two report chapters is greater than a preset threshold; (3) the two report chapters have a business logic relationship, such as belonging to the same business process. When any of the above conditions are met, an edge is constructed between the two nodes to ensure that the construction of the edge conforms to the actual business logic.

[0098] Step S2013: Construct a graph structure based on the nodes corresponding to each report chapter in the target report, the node characteristics of each node, and the edges between nodes.

[0099] Step S2014: Obtain multiple sample report chapters, construct a sample graph structure based on the multiple sample report chapters, fine-tune the pre-trained graph neural network based on the sample graph structure, and obtain the target graph neural network. Each sample report chapter carries sample annotations.

[0100] Specifically, each sample report section carries a corresponding sample label, indicating whether the sample report section should be merged and its relationship with other sample report sections. A sample graph structure is constructed based on these sample report sections, where each node represents a sample report section. Node features are extracted and edges between nodes are constructed, referring to steps S2011 and S2012. Graph Neural Networks (GNNs) have been pre-trained on a general dataset and possess basic graph structure feature learning capabilities. To better adapt it to specific scenarios of enterprise report section association analysis (such as accurate identification of section references and business logic associations), fine-tuning is required through the sample graph structure. Using nodes and their features as input, a two-layer GNN model structure is adopted, combining cross-entropy loss and graph structure regularization loss functions. The model is iteratively trained using sample labels as supervision signals to optimize model parameters, ultimately obtaining a target graph neural network adapted to the current enterprise report section association analysis scenario. This improves the model's learning accuracy of section references, semantics, and business associations, providing more practical model support for subsequent report graph association mining.

[0101] Step S2015: Using a target graph neural network, update the node features of each node in the graph structure based on the node features of its neighboring nodes.

[0102] Specifically, the target GNN obtained by fine-tuning is used to perform a fusion calculation on each node in the graph structure based on the node features of the neighboring nodes directly connected to it through edges, and the node features of the current node itself. This realizes the update of node features, so that the node features of each node not only retain their own business attributes, but also incorporate the related context of neighboring nodes, further improving the ability of node features to represent the business relationship of chapters.

[0103] Step S2016: Based on each updated node in the graph structure, obtain the report graph.

[0104] Step S202: Determine the business type of each node in the report graph, and based on the business types of all nodes, divide them into multiple similar to-do groups.

[0105] In some optional implementations, step S202 above determines the service type of each node in the report graph, including:

[0106] Step S2021: Using a natural language processing classifier, the initial type of a node is determined based on the node features of each node in the report graph.

[0107] Specifically, the Natural Language Processing Classifier (NLP) has been fine-tuned using a pre-trained BERT model. During actual judgment, the classifier takes the node features of each node in the report graph as input, focusing on extracting core information reflecting the chapter's business attributes from the node features, such as chapter title, chapter content, and keywords. Through the model's understanding and analysis of text semantics, it initially predicts the business type of the report chapter corresponding to the node.

[0108] Step S2022: In the business rule engine, call the rule corresponding to the initial type to verify the keywords and chapter content in the node characteristics of the node.

[0109] Specifically, the business rules engine has built-in business rules for different business types. It calls the business rules corresponding to the initial type to validate the keywords and content of each report section. For example, taking a node with an initial type of "financial approval" (such as the "department budget preparation" section in an annual budget report), it checks whether the keyword contains the word "budget" and whether the content contains any statements related to "amount." Optionally, the validation process supports fuzzy matching to be compatible with keyword variations, and the amount recognition also covers multiple currency formats, further improving the flexibility and accuracy of rule validation.

[0110] Step S2023: When the keywords and chapter content of the node have both passed the verification, the initial type is determined as the business type of the node.

[0111] Specifically, when both keywords and chapter content pass validation, it indicates that the initial type fully matches the business attributes of the report chapter corresponding to the node, and thus can be used as the business type of the report chapter. This dual validation improves the accuracy of business type classification, providing support for subsequent chapter aggregation.

[0112] Step S2024: If the keyword or chapter content verification of a node fails, the initial type is manually reviewed to obtain the business type of the node.

[0113] Specifically, when keywords or chapter content fail the verification, it indicates a discrepancy between the initial type and the actual business attributes of the chapter, triggering a manual review process to re-determine the business type of the node and ensure the accuracy of the business type determination.

[0114] Step S203: For each group of tasks of the same type, based on the node characteristics of each node, calculate the quantum entanglement degree between any two nodes in the group of tasks of the same type to obtain the semantic entanglement degree of each node.

[0115] Specifically, step S203 includes:

[0116] Step S2031: For any two nodes in each to-do group of the same type, calculate the semantic similarity, workflow similarity, time comparison value and permission conflict value between the two nodes based on the node characteristics of the two nodes.

[0117] Specifically, step S2031 includes:

[0118] Step a1: For any two nodes in each to-do group of the same type, determine the semantic similarity between the two nodes based on the chapter title and keywords in the node features of the two nodes.

[0119] Specifically, for any two nodes within the same group, after converting the chapter titles of the two nodes into vectors, the cosine similarity between the chapter titles is calculated. Then, the overlap of keywords between the two nodes is calculated. For example, if the keywords of node A are "budget" and "finance," and the keywords of node B are "funding," "finance," and "planning," the keyword overlap is 1 / 5 = 0.2. Assuming the weight corresponding to the chapter title is 0.6 and the weight corresponding to the keyword is 0.4, a weighted sum of these two similarities is taken to obtain the semantic similarity between the two nodes.

[0120] Step a2: Based on the approval paths in the node features of the two nodes, identify the common approval nodes and all approval nodes in the approval paths of the two nodes, and determine the ratio of the common approval nodes to all approval nodes as the workflow similarity between the two nodes.

[0121] Specifically, assuming the approval path of node A is department head → financial director → general manager, and the approval path of node B is department head → legal director → general manager, we filter out the common approval nodes of the two: department head and general manager. After removing duplicate nodes, we count all approval nodes: department head, financial director, legal director and general manager. Thus, the workflow similarity is 2 / 4 = 0.5, which quantifies the degree of overlap between the two nodes in the approval process.

[0122] Step a3: Calculate the difference in deadlines between the two nodes based on their node characteristics, and calculate the time comparison value between the two nodes based on the difference in deadlines and a preset time interval.

[0123] Specifically, assume that the deadline for node A is September 10, 2025, and the deadline for node B is September 12, 2025, with a difference of 2 days. The preset time interval is the maximum time difference threshold allowed by the system, with a default configuration of 7 days. If the time difference exceeds 7 days, the two nodes are considered to be unrelated. Used to measure the degree of coordination between two nodes over time.

[0124] Step a4: Determine the permission conflict value between the two nodes based on the approval path in the node characteristics of the two nodes.

[0125] Specifically, based on the approval paths of two nodes, the permission graph is used to determine whether there are permission conflicts among the approval nodes in the path. If the two nodes belong to the same department (e.g., both approval paths involve the finance department), there is no conflict, and the permission conflict value is 0. If the approval nodes belong to different departments (e.g., node A involves the finance department, and node B involves the legal department), the permission conflict value is 1. The permission graph is constructed based on the company's organizational structure data. Nodes include users, roles, permissions, and departments, and the relationships between nodes include affiliation, ownership, inheritance, and mutual exclusion. Permission change events are monitored in real time, and the permission graph is updated incrementally.

[0126] Step S2032: Based on the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes and their corresponding weights, a weighted calculation is performed to obtain the quantum entanglement degree between the two nodes.

[0127] Specifically, the concept of quantum entanglement used in this embodiment of the invention does not represent the realization of a physical quantum system. Instead, it uses analogical logic and borrows quantum physics models to solve complex correlation analysis problems in the field of information processing. More specifically, quantum entanglement is a metaphorical borrowing and mathematical abstraction of a concept from quantum physics, concretely and quantitatively describing the inseparable and mutually influential strong correlation between two report sections. It does not represent the quantum entanglement phenomenon of microscopic particles in a macroscopic computer system.

[0128] Weights are preset for the four dimensions mentioned above, with the sum of the weights being 1. Semantic similarity, workflow similarity, and time comparison are all positive indicators; the larger the value, the more relevant the nodes are. Permission conflict is a negative indicator; the larger the value, the more severe the conflict and the less relevant the nodes are. The quantum entanglement degree between two nodes is obtained by weighting the preset weights with the values ​​of each dimension using the following formula (1). Formula (1) uses subtraction logic to handle negative indicators, which aligns with the idea of ​​indivisibility in quantum entanglement, meaning that any mismatch in any dimension (such as permission conflict) will significantly affect the overall association state. The larger the final calculated quantum entanglement degree, the more inseparable the two report sections are in terms of business, like entangled particles, requiring treatment as a whole. By comprehensively covering the characteristics of report sections in semantic, workflow, time, and permission association dimensions, the quantum entanglement degree accurately reflects the business association between nodes.

[0129] Quantum entanglement degree = w1 × semantic similarity + w2 × workflow similarity + w3 × temporal coordination degree - w4 × permission conflict value (1)

[0130] In the formula, w1, w2, w3, and w4 represent the weights of semantic similarity, workflow similarity, time comparison value, and permission conflict value, respectively.

[0131] Step S2033: For each node, take the maximum value among all quantum entanglement degrees corresponding to the node as the semantic entanglement degree of the node.

[0132] Specifically, each node calculates its quantum entanglement degree with all other nodes in the group one by one. To highlight the strongest correlation between a node and other nodes in the group, the maximum value among these quantum entanglement degrees is taken as the semantic entanglement degree of that node.

[0133] Step S204: When the semantic entanglement degree of a node is greater than a preset threshold, calculate the node's merging score, and based on the merging score, determine whether the node should be merged or split.

[0134] Specifically, step S204 includes:

[0135] Step S2041: When a node is an isolated node in the report graph, set the value of the node in the business continuity dimension to a first preset value; when a node is not an isolated node in the report graph, set the value of the node in the business continuity dimension to a second preset value.

[0136] Specifically, when a node has no connecting edges in the report graph, that is, the node is an isolated node, the value of the node in the business continuity dimension is set to 0, which means that the report chapter is independent and does not need to rely on the previous chapter for processing; when a node has connecting edges in the report graph, that is, the node is not an isolated node, the value of the node in the business continuity dimension is set to 1, which means that the report chapter has a dependency relationship and needs to rely on the previous related chapter for processing.

[0137] In step S2042, when the node's node characteristics do not include a timeliness identifier and the node's business type is not timeliness sensitive, the node's time sensitivity dimension is set to a first preset value. When the node's node characteristics include a timeliness identifier or the node's business type is timeliness sensitive, the node's time sensitivity dimension is set to a second preset value.

[0138] Specifically, if a node's node characteristics do not include time-sensitive markers such as "complete within 3 days" or "urgent approval," and its business type does not belong to time-sensitive types explicitly stated in documents such as "financial audit" or "compliance review," then the time sensitivity dimension is assigned a value of 0. If a node's node characteristics include time-sensitive markers, or its business type belongs to time-sensitive types, then the time sensitivity dimension is assigned a value of 1, ensuring that the assignment matches the actual time constraint requirements of the chapter.

[0139] Step S2043: When the node features include a single approval node, set the node's permission complexity dimension to a first preset value; when the node features include multiple approval nodes, set the node's permission complexity dimension to a second preset value.

[0140] Specifically, if a node's node characteristics only include a single approval node, the permission complexity dimension is assigned a value of 0, representing single-role processing and simple permission requirements; if a node's node characteristics include multiple approval nodes, the permission complexity dimension is assigned a value of 1, representing multi-role collaboration and complex permission requirements.

[0141] Step S2044: Based on the values ​​assigned to the nodes in the business continuity dimension, time sensitivity dimension, and permission complexity dimension, obtain the multi-dimensional merge matrix corresponding to the nodes.

[0142] Specifically, for each node, its values ​​in the dimensions of business continuity, time sensitivity, and permission complexity are integrated in the order of business continuity → time sensitivity → permission complexity to form a multi-dimensional merge matrix (a 1×3 row vector), such as [0,1,0], [0,1,1], etc. By constructing the multi-dimensional merge matrix, the business, time, and permission attributes of the nodes can be accurately quantified.

[0143] Step S2045: The values ​​assigned to each dimension in the multi-dimensional merging matrix and the preset dimension weights are weighted and summed to obtain the merging score of the node.

[0144] Specifically, preset dimension weights are configured for the three dimensions, such as 0.4 for business continuity, 0.3 for time sensitivity, and 0.3 for permission complexity. The value of each dimension in the multi-dimensional merging matrix is ​​multiplied by its corresponding weight and then summed to obtain the merging score of the node. For example, the merging score of the matrix [1,1,1] is 1 × 0.4 + 1 × 0.3 + 1 × 0.3 = 1. Determining the merging score provides a basis for judging the refined aggregation of chapters.

[0145] Step S2046: Obtain the preset merging probability threshold.

[0146] Specifically, the preset merge probability threshold is not a fixed value, but is dynamically adjusted according to the system's operating status and business scenarios to adapt to the needs of aggregation efficiency and security compliance in different scenarios. When the system is busy, such as when CPU utilization or memory usage exceeds 80%, the preset merge probability threshold is set to 0.6 to reduce the number of pending tasks and reduce system processing pressure. A lower threshold allows more nodes to meet the merge conditions and is more likely to trigger aggregation. When a report section involves highly sensitive operations, the threshold is set to 0.8 to avoid unauthorized merging and ensure permission security. A higher threshold raises the merging threshold and makes it more difficult to trigger aggregation.

[0147] Step S2047: Calculate the merging probability of a node based on its merging score and a preset merging probability threshold.

[0148] Specifically, the merging probability is calculated using the following formula (2), which transforms the discrete merging score into a continuous probability value, thus more accurately reflecting the possibility of node merging.

[0149]

[0150] In the formula, ε represents the merging probability; k represents the adjustment coefficient. The larger k is, the more sensitive the merging probability is to the merging score; the smaller k is, the more the merging score is allowed to fluctuate around the preset merging probability threshold. Generally, k is between 4 and 8 to balance sensitivity and stability. If the approval node of the report chapter has high authority, K can be set to 10 to indicate strict decision-making; η represents the merging score; and μ represents the preset merging probability threshold.

[0151] Step S2048: Calculate the merging tendency coefficient and splitting tendency coefficient based on the merging probability of the nodes.

[0152] Specifically, the merging tendency coefficient and splitting tendency coefficient are calculated by the following formula (3) to quantify the tendency of nodes to transform into merging and splitting states.

[0153]

[0154] In the formula, α represents the merging tendency coefficient; ε represents the merging probability; and β represents the splitting tendency coefficient.

[0155] Step S2049: Generate a random value within a preset interval. When the random value is less than the merging tendency coefficient or the merging tendency coefficient is greater than the splitting tendency coefficient, determine the nodes to be merged.

[0156] Specifically, the preset interval is [0,1]. A value is randomly generated within this interval. If the value is less than the merging tendency coefficient or the merging tendency coefficient is greater than the splitting tendency coefficient, then the report chapter corresponding to the node is considered to be aggregated into the total to-do tasks of the same type of to-do group.

[0157] Step S20410: When the random value is greater than the merging tendency coefficient or the merging tendency coefficient is not greater than the splitting tendency coefficient, determine the node to be split.

[0158] Specifically, if the value is greater than the merging tendency coefficient or the merging tendency coefficient is not greater than the splitting tendency coefficient, then the report chapter corresponding to the node is considered to need to be split and then aggregated into the total to-do tasks.

[0159] Step S205: When splitting a node, perform conflict detection on the node and similar task groups to obtain the conflict type. See details below. Figure 1Step S105 of the illustrated embodiment will not be described again here.

[0160] Step S206: According to the splitting method corresponding to the conflict type, split similar to-do groups into multiple split to-do groups. Generate sub-to-do tasks for each split to-do group within the total to-do tasks corresponding to the similar to-do groups. Aggregate the report sections corresponding to the nodes into the sub-to-do tasks of the corresponding split to-do groups. For details, please refer to... Figure 1 Step S106 of the illustrated embodiment will not be described again here.

[0161] Step S207: When merging nodes, determine whether the nodes have compatible permissions with similar to-do groups.

[0162] Specifically, when a node is determined to need to be merged, a permission compatibility check is performed on that node and all other nodes in the same pending group according to the permission graph. During the check, the approval path of the current node and a certain node in the group is extracted. If the approval nodes of the two paths are the same or there is no permission conflict between the different approval nodes, the permissions are determined to be compatible; if there is a permission conflict between the different approval nodes, the permissions are determined to be incompatible.

[0163] Step S208: When the node's permissions are compatible with those of the same type of to-do group, aggregate the report section corresponding to the node into the total to-do task corresponding to the same type of to-do group.

[0164] Specifically, if the permissions of the node are compatible with those of all nodes in the group, the report section of the node will be directly aggregated into the total task of the same type of task group, as a sub-item of the total task.

[0165] Step S209: When the permissions of a node are incompatible with any node in the same type of to-do group, the same type of to-do group is split into multiple split to-do groups according to permissions. Sub-to-do tasks corresponding to each split to-do group are generated in the total to-do tasks corresponding to the same type of to-do group. The report chapters corresponding to the node are aggregated into the sub-to-do tasks of the corresponding split to-do group.

[0166] Specifically, if it is detected that the current node's permissions are incompatible with any node in the same type of to-do group, the to-do group is split according to permissions: nodes with consistent permissions are grouped into the same split to-do group. Under the main to-do task corresponding to the same type of to-do group, sub-to-do tasks are generated for each split to-do group, and the report chapter corresponding to the current node is precisely aggregated into the sub-to-do task that matches its permissions, achieving fine-grained aggregation of chapters.

[0167] In some alternative implementations, assuming the target report is the Q3 2025 budget report, the to-do list generated by the reporting system would look like this:

[0168] Financial Audit (Total Tasks) (7)

[0169] {

[0170] Finance Department Approval (Sub-tasks to be done) (3 sub-items)

[0171] Ministry of Justice Approval (Sub-tasks to be completed) (2 sub-items)

[0172] Chapter 4

[0173] Chapter 7

[0174] }

[0175] Legal Compliance Tasks (Total Tasks) (3)

[0176] {

[0177] Data Security Compliance Report - Appendix A

[0178] (The remaining two items are folded)

[0179] }

[0180] Optionally, the total task list, sub-tasks within the total task list, sub-tasks, and sub-sub-tasks within sub-tasks can be sorted according to the urgency of the report section.

[0181] Step S210: When the target report is detected to have been published, delete all pending tasks corresponding to the target report.

[0182] Specifically, by monitoring the entire lifecycle of report status changes through an event bus (such as Kafka), when the system detects that a target report has triggered a "publish" event—that is, when the report has completed all approval processes and is officially published—a pending task cleanup mechanism is automatically triggered. This mechanism first identifies all pending tasks corresponding to the target report using its unique identifier, and then deletes these associated pending tasks in batches without manual intervention. This avoids redundant pending tasks from published reports consuming system resources or interfering with user operations, and also supports rollback in case of exceptions.

[0183] This invention transforms the target report into a report graph, explicitly displaying the chapter information and relationships between chapters. It determines the business type of each node in the graph and categorizes them into similar task groups, achieving initial chapter aggregation. Targeted analysis is then performed on each similar task group. For each node within the group, the semantic entanglement degree is calculated based on its correlation with other nodes in the group. Based on this initial aggregation, nodes with high correlation are quickly selected according to the semantic entanglement degree and a preset threshold, achieving coarse screening of node associations. A merging score is calculated for the coarsely screened nodes, and the merging score determines whether to merge or split the nodes, enabling refined decision-making. Conflict detection is performed during node splitting, and conflict types are determined. Sub-tasks are generated based on conflict types, reducing the number of tasks while ensuring compliance. Initial aggregation is achieved through similar task group division, and further aggregation is performed using semantic entanglement, reducing the number of generated tasks and accurately aggregating each report chapter into appropriate tasks, achieving refined aggregation and effectively reducing the user's operational burden and time cost.

[0184] This embodiment also provides a chapter task aggregation device based on semantic entanglement, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0185] This embodiment provides a chapter task aggregation device based on semantic entanglement, such as... Figure 2 As shown, it includes:

[0186] The generation module 201 is used to generate a report graph based on the target report. The report graph includes multiple nodes and their node features. The nodes represent report chapters in the target report.

[0187] The determination module 202 is used to determine the business type of each node in the report graph, and based on the business types of all nodes, to divide them into multiple similar to-do groups.

[0188] The calculation module 203 is used to calculate the quantum entanglement degree between any two nodes in each type of task group based on the node characteristics of each node, and to obtain the semantic entanglement degree of each node.

[0189] The first judgment module 204 is used to calculate the merging score of a node when the semantic entanglement degree of the node is greater than a preset threshold, and to determine whether the node should be merged or split based on the merging score.

[0190] The detection module 205 is used to perform conflict detection on nodes and similar task groups when nodes are split, and to obtain the conflict type.

[0191] The first aggregation module 206 is used to split similar to-do groups into multiple split to-do groups according to the splitting method corresponding to the conflict type, generate sub-to-do tasks corresponding to each split to-do group in the total to-do tasks corresponding to the similar to-do groups, and aggregate the report chapters corresponding to the nodes into the sub-to-do tasks of the corresponding split to-do groups.

[0192] In some optional embodiments, after the first determination module 204, the device further includes:

[0193] The second judgment module is used to determine whether a node has compatible permissions with other nodes in the same type of pending group when merging nodes.

[0194] The second aggregation module is used to aggregate the report chapters corresponding to a node into the total to-do tasks corresponding to the same to-do group when the node's permissions are compatible with those of the same to-do group.

[0195] The third aggregation module is used to split the same type of to-do group into multiple split to-do groups according to permissions when the permissions of a node are incompatible with any node in the same type of to-do group. It generates sub-to-do tasks corresponding to each split to-do group in the total to-do tasks corresponding to the same type of to-do group, and aggregates the report chapters corresponding to the node into the sub-to-do tasks of the corresponding split to-do group.

[0196] In some optional implementations, the generation module 201 includes:

[0197] The first construction unit is used to take each report chapter in the target report as a node, extract the chapter features of each report chapter, and use them as the node features of the corresponding node of the report chapter.

[0198] The second building unit is used to build an edge between two nodes when there is a reference relationship between them, and / or the semantic similarity between them is greater than a preset threshold, and / or there is a business logic relationship between them.

[0199] The third building unit is used to construct a graph structure based on the nodes corresponding to each report chapter in the target report, the node characteristics of each node, and the edges between nodes.

[0200] The fine-tuning unit is used to acquire multiple sample report chapters, construct a sample graph structure based on the multiple sample report chapters, fine-tune the pre-trained graph neural network based on the sample graph structure, and obtain the target graph neural network. Each sample report chapter carries sample annotations.

[0201] The update unit is used to employ a target graph neural network to update the node features of each node in the graph structure based on the node features of the node's neighboring nodes.

[0202] The generation unit is used to generate a report graph based on each updated node in the graph structure.

[0203] In some alternative implementations, the determining module 202 includes:

[0204] The first determining unit is used to determine the initial type of a node based on the node features of each node in the report graph using a natural language processing classifier.

[0205] The verification unit is used in the business rule engine to call the rules corresponding to the initial type and verify the keywords and chapter content in the node characteristics of the node.

[0206] The second determining unit is used to determine the initial type as the business type of the node when both the keywords and chapter content of the node pass the verification.

[0207] The third determining unit is used to manually review the initial type when the keyword or chapter content verification of a node fails, in order to obtain the business type of the node.

[0208] In some alternative implementations, the computing module 203 includes:

[0209] The first calculation unit is used to calculate the semantic similarity, workflow similarity, time comparison value, and permission conflict value between any two nodes in each to-do group of the same type, based on the node characteristics of the two nodes.

[0210] The second computing unit is used to perform weighted calculations based on the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes and their corresponding weights to obtain the quantum entanglement degree between the two nodes.

[0211] The fourth determining unit is used to take the maximum value among all quantum entanglement degrees corresponding to each node as the semantic entanglement degree of the node.

[0212] In some alternative implementations, the first computing unit includes:

[0213] The first determining subunit is used to determine the semantic similarity between any two nodes in each to-do group of the same type, based on the chapter title and keywords in the node features of the two nodes.

[0214] The second determining subunit is used to determine the same approval nodes and all approval nodes in the approval paths of the two nodes based on the node features of the two nodes, and to determine the ratio of the same approval nodes to all approval nodes as the workflow similarity between the two nodes.

[0215] The third determining sub-unit is used to calculate the difference in deadlines between two nodes based on the deadlines in the node characteristics of the two nodes, and to calculate the time comparison value between the two nodes based on the difference in deadlines and a preset time interval.

[0216] The fourth determination subunit is used to determine the permission conflict value between two nodes based on the approval path in the node characteristics of the two nodes.

[0217] In some optional implementations, the first determination module 204 includes:

[0218] The first setting unit is used to set the value of the node in the business continuity dimension to a first preset value when the node is an isolated node in the report graph, and to set the value of the node in the business continuity dimension to a second preset value when the node is not an isolated node in the report graph.

[0219] The second setting unit is used to set the node's assignment value in the time sensitivity dimension to a first preset value when the node's node characteristics do not include a timeliness identifier and the node's business type is not timeliness sensitive, and to set the node's assignment value in the time sensitivity dimension to a second preset value when the node's node characteristics include a timeliness identifier or the node's business type is timeliness sensitive.

[0220] The third setting unit is used to set the node's permission complexity dimension to a first preset value when the node's node characteristics include a single approval node, and to set the node's permission complexity dimension to a second preset value when the node's node characteristics include multiple approval nodes.

[0221] The fifth determining unit is used to obtain the multi-dimensional merging matrix corresponding to the node based on the values ​​assigned to the node in the dimensions of business continuity, time sensitivity, and permission complexity.

[0222] The sixth determining unit is used to perform a weighted summation of the assigned values ​​of each dimension in the multi-dimensional merging matrix and the preset dimension weights to obtain the merging score of the node.

[0223] In some optional implementations, the first determination module 204 includes:

[0224] The acquisition unit is used to acquire a preset merging probability threshold.

[0225] The third calculation unit is used to calculate the merging probability of nodes based on the node merging score and the preset merging probability threshold.

[0226] The fourth calculation unit is used to calculate the merging tendency coefficient and splitting tendency coefficient based on the merging probability of nodes.

[0227] The seventh determining unit is used to generate random values ​​within a preset interval. When the random value is less than the merging tendency coefficient or the merging tendency coefficient is greater than the splitting tendency coefficient, the node is determined to be merged.

[0228] The eighth determining unit is used to determine the node to be split when the random value is greater than the merging tendency coefficient or the merging tendency coefficient is not greater than the splitting tendency coefficient.

[0229] In some alternative embodiments, the device further includes:

[0230] The delete module is used to delete all pending tasks corresponding to a target report when the target report has been published.

[0231] 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.

[0232] In this embodiment, the chapter task aggregation device based on semantic entanglement is presented in the form of a functional unit. 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.

[0233] This invention also provides a computer device having the above-described features. Figure 2 The illustrated chapter task aggregation device is based on semantic entanglement.

[0234] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 3 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 3 Take a processor 10 as an example.

[0235] 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.

[0236] 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.

[0237] 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.

[0238] 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.

[0239] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0240] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0241] 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.

[0242] 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.

[0243] 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 chapter task aggregation method based on semantic entanglement, characterized in that, The method includes: A report graph is generated based on the target report. The report graph includes multiple nodes and their node features. The nodes represent report sections in the target report. The business type of each node in the report graph is determined, and multiple similar to-do groups are obtained based on the business types of all nodes; For each group of tasks of the same type, based on the node characteristics of each node, the quantum entanglement degree between any two nodes in the group of tasks of the same type is calculated to obtain the semantic entanglement degree of each node. When the semantic entanglement degree of a node is greater than a preset threshold, the node's merging score is calculated, and based on the merging score, it is determined whether the node should be merged or split. When the node is split, conflict detection is performed on the node and the same type of to-do group to obtain the conflict type; According to the splitting method corresponding to the conflict type, the same type of to-do group is split into multiple split to-do groups. Sub-to-do tasks corresponding to each split to-do group are generated in the total to-do tasks corresponding to the same type of to-do group. The report chapters corresponding to the node are aggregated into the sub-to-do tasks of the corresponding split to-do group. Specifically, for each similar task group, based on the node characteristics of each node, the quantum entanglement degree between any two nodes in the similar task group is calculated to obtain the semantic entanglement degree of each node, including: For any two nodes in each to-do group of the same type, based on the node characteristics of the two nodes, calculate the semantic similarity, workflow similarity, time comparison value and permission conflict value between the two nodes; The quantum entanglement degree between the two nodes is obtained by weighting the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes and their corresponding weights. For each node, the maximum value among all quantum entanglement degrees corresponding to the node is taken as the semantic entanglement degree of the node; For any two nodes in each similar to-do group, based on the node characteristics of the two nodes, the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes are calculated, including: For any two nodes in each to-do group of the same type, the semantic similarity between the two nodes is determined based on the chapter title and keywords in the node features of the two nodes; Based on the approval paths in the node features of the two nodes, the same approval nodes and all approval nodes in the approval paths of the two nodes are determined, and the ratio of the same approval nodes to all approval nodes is determined as the workflow similarity between the two nodes. Based on the deadline times in the node characteristics of the two nodes, calculate the difference between the deadline times of the two nodes, and based on the difference between the deadline times and a preset time interval, calculate the time comparison value between the two nodes. Based on the approval path in the node characteristics of the two nodes, the permission conflict value between the two nodes is determined.

2. The method according to claim 1, characterized in that, After determining whether the nodes are merged or split, the method further includes: When merging nodes, determine whether the node has compatible permissions with the same type of to-do group; When the node is compatible with the permissions of the same type of to-do group, the report section corresponding to the node is aggregated into the total to-do task corresponding to the same type of to-do group; When the permissions of the node are incompatible with any node in the same type of to-do group, the same type of to-do group is split into multiple split to-do groups according to permissions. Sub-to-do tasks corresponding to each split to-do group are generated in the total to-do tasks corresponding to the same type of to-do group. The report chapters corresponding to the node are aggregated into the sub-to-do tasks of the corresponding split to-do group.

3. The method according to claim 1, characterized in that, The process of generating a report map based on the target report includes: Each report section in the target report is taken as a node, and the section features of each report section are extracted as the node features of the corresponding node of the report section. When there is a reference relationship between any two nodes, and / or the semantic similarity between the two nodes is greater than a preset threshold, and / or there is a business logic association between the two nodes, an edge is constructed between the two nodes. A graph structure is constructed based on the nodes corresponding to each report chapter in the target report, the node characteristics of each node, and the edges between the nodes; Multiple sample report chapters are obtained, and a sample graph structure is constructed based on the multiple sample report chapters. The pre-trained graph neural network is fine-tuned based on the sample graph structure to obtain the target graph neural network. Each sample report chapter carries sample annotations. Using the target graph neural network, for each node in the graph structure, the node features are updated based on the node features of the node's neighboring nodes; The report graph is obtained based on each updated node in the graph structure.

4. The method according to claim 1, characterized in that, Determining the service type of each node in the report graph includes: A natural language processing classifier is used to determine the initial type of each node based on the node features of each node in the report graph; In the business rule engine, the rule corresponding to the initial type is called to verify the keywords and chapter content in the node features of the node respectively; When both the keywords and chapter content of the node pass the verification, the initial type is determined as the business type of the node; If the keyword or chapter content verification of the node fails, the initial type is manually reviewed to obtain the business type of the node.

5. The method according to claim 1, characterized in that, The calculation of the merged score of the node includes: When the node is an isolated node in the report graph, the value assigned to the node in the business continuity dimension is set to a first preset value; when the node is not an isolated node in the report graph, the value assigned to the node in the business continuity dimension is set to a second preset value. When the node's node features do not include a timeliness identifier and the node's business type is not timeliness sensitive, the node's assignment value in the time sensitivity dimension is set to the first preset value. When the node's node features include the timeliness identifier or the node's business type is timeliness sensitive, the node's assignment value in the time sensitivity dimension is set to the second preset value. When the node features include a single approval node, the node is assigned a value of the first preset value in the permission complexity dimension; when the node features include multiple approval nodes, the node is assigned a value of the second preset value in the permission complexity dimension. Based on the values ​​assigned to the node in the business continuity dimension, the time sensitivity dimension, and the permission complexity dimension, a multi-dimensional merging matrix corresponding to the node is obtained; The combined score of the node is obtained by weighting and summing the values ​​of each dimension in the multi-dimensional merging matrix and the preset dimension weights.

6. The method according to claim 1, characterized in that, The step of determining whether to merge or split the node based on the merge score includes: Obtain the preset merging probability threshold; The merging probability of the node is calculated based on the node's merging score and the preset merging probability threshold. Based on the merging probability of the nodes, calculate the merging tendency coefficient and the splitting tendency coefficient; A random value is generated within a preset range. When the random value is less than the merging tendency coefficient or the merging tendency coefficient is greater than the splitting tendency coefficient, the node is determined to be merged. When the random value is greater than the merging tendency coefficient or the merging tendency coefficient is not greater than the splitting tendency coefficient, the node is determined to be split.

7. The method according to claim 1, characterized in that, The method further includes: Once the target report is detected to have been published, delete all pending tasks corresponding to the target report.

8. A chapter task aggregation device based on semantic entanglement, characterized in that, The device includes: The generation module is used to generate a report graph based on the target report. The report graph includes multiple nodes and their node features, and the nodes represent report sections in the target report. The determination module is used to determine the business type of each node in the report graph, and to divide the data into multiple similar to-do groups based on the business types of all nodes; The calculation module is used to calculate the quantum entanglement degree between any two nodes in each similar task group based on the node characteristics of each node, and to obtain the semantic entanglement degree of each node. The first judgment module is used to calculate the merging score of the node when the semantic entanglement degree of the node is greater than a preset threshold, and to determine whether the node should be merged or split based on the merging score. The detection module is used to perform conflict detection on the node and the same type of to-do group when the node is split, and to obtain the conflict type; The first aggregation module is used to split the same type of task group into multiple split task groups according to the splitting method corresponding to the conflict type, generate sub-tasks corresponding to each split task group in the total task group corresponding to the same type of task group, and aggregate the report chapters corresponding to the node into the sub-tasks of the corresponding split task group. Specifically, the calculation module is used for: For any two nodes in each to-do group of the same type, based on the node characteristics of the two nodes, calculate the semantic similarity, workflow similarity, time comparison value and permission conflict value between the two nodes; The quantum entanglement degree between the two nodes is obtained by weighting the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes and their corresponding weights. For each node, the maximum value among all quantum entanglement degrees corresponding to the node is taken as the semantic entanglement degree of the node; For any two nodes in each similar to-do group, based on the node characteristics of the two nodes, the semantic similarity, workflow similarity, time comparison value, and permission conflict value between the two nodes are calculated, including: For any two nodes in each to-do group of the same type, the semantic similarity between the two nodes is determined based on the chapter title and keywords in the node features of the two nodes; Based on the approval paths in the node features of the two nodes, the same approval nodes and all approval nodes in the approval paths of the two nodes are determined, and the ratio of the same approval nodes to all approval nodes is determined as the workflow similarity between the two nodes. Based on the deadline times in the node characteristics of the two nodes, calculate the difference between the deadline times of the two nodes, and based on the difference between the deadline times and a preset time interval, calculate the time comparison value between the two nodes. Based on the approval path in the node characteristics of the two nodes, the permission conflict value between the two nodes is determined.

Images