Target text generation method based on big data

Abstract

The present invention relates to the technical field of artificial intelligence and natural language processing, and discloses a target text generation method based on big data, including: modeling the document structure as a directed graph, using a graph attention network to predict the optimal document structure; constructing a document association graph and using a graph neural network to analyze the relationship between documents; constructing a document dependency graph to form a complete document knowledge network; based on three graph structures, using a graph propagation algorithm to generate text content; when the source data changes, automatically triggering the graph propagation algorithm to perform intelligent updates; the present invention breaks through the limitations of traditional templated structures, realizes adaptive document organization, accurately identifies complex associations between documents, forms an organic knowledge network, realizes intelligent propagation of document updates, automatically maintains dependencies and version consistency, and improves the quality and efficiency of document generation in the consulting, design and planning industries.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and natural language processing, and more particularly, to a target text generation method based on big data. Background Art

[0002] With the rapid development of artificial intelligence and big data technologies, data-driven intelligent text generation technology has been widely used in many fields. However, in professional fields such as consulting, design and planning, current text generation technology still faces many challenges.

[0003] Traditional text generation methods rely primarily on pre-set templates and rules, lacking a deep understanding of content semantics and flexible organizational capabilities. These methods typically employ techniques such as keyword matching and sentence substitution, making them ineffective for complex professional document generation. The limitations of traditional methods are particularly evident when it comes to comprehensively analyzing large amounts of data and generating logically sound, highly specialized documents.

[0004] When it comes to document structure, existing technologies often use fixed templates or simple statistical methods to determine the document structure, failing to intelligently adjust it based on specific content and usage scenarios. This results in a rigid document structure that is difficult to adapt to the specific needs of different document types, affecting the overall quality and practical value of the document.

[0005] Furthermore, existing technologies also have significant shortcomings in handling document updates and version management. When underlying data or related documents change, existing systems struggle to automatically identify and update all affected documents, requiring extensive manual intervention, which is inefficient and prone to errors.

[0006] Therefore, a target text generation method is needed that can intelligently understand the document structure, accurately identify the relationship between documents, and support automatic updates to meet the needs of high-quality document generation in professional fields such as consulting, design and planning. Summary of the Invention

[0007] The present invention provides a target text generation method based on big data, which solves the technical problems in related technologies such as the lack of deep understanding of content semantics and flexible organization capabilities, and the inability to effectively handle complex professional document generation needs.

[0008] The present invention provides a target text generation method based on big data, comprising the following steps:

[0009] The structure of the target document is modeled as a directed graph, and the graph attention network is used to analyze the document structure graph. The importance weight of each node is calculated through the attention algorithm to predict the optimal document structure organization method.

[0010] Based on the predicted document structure, a document association graph is constructed. Graph neural networks are used to analyze the semantic similarity and logical association strength between documents to identify implicit relationships between documents.

[0011] Based on the document association graph, a document dependency graph is constructed to clarify the dependency and version correspondence between documents, thus forming a complete document knowledge network structure.

[0012] Based on the constructed document structure graph, document association graph, and dependency graph, the graph propagation algorithm is used to generate text content;

[0013] When the source data or related documents change, the graph propagation algorithm is automatically triggered to intelligently update the structure and content of the affected documents based on the document dependencies.

[0014] In a preferred embodiment, the graph attention network adopts a multi-layer cascade structure, including 3 to 5 attention layers, each layer using 8 to 16 attention heads, and the graph attention network includes a multi-head attention layer, a feature transformation layer and an output layer.

[0015] In a preferred embodiment, the graph neural network includes a graph convolution layer, a pooling layer, and a classification layer. The graph convolution layer is responsible for feature propagation and aggregation on the document graph, the pooling layer summarizes the graph-level features, and the classification layer outputs the association strength score between documents.

[0016] In a preferred embodiment, the document dependency graph includes a dependency relationship identification module, a version management module, and a consistency checking module.

[0017] In a preferred embodiment, the graph propagation algorithm adopts a multi-round iterative propagation method and gradually converges to a stable state through 3 to 10 rounds of iterations. The graph propagation algorithm includes an information aggregation unit, a weight calculation unit and a content generation unit.

[0018] In a preferred embodiment, the propagation weight is dynamically calculated based on the semantic similarity, citation frequency and temporal correlation between documents.

[0019] In a preferred embodiment, the dynamic update adopts an incremental update strategy, and only the affected document portion is updated.

[0020] In a preferred embodiment, the graph attention network uses residual connections and layer normalization techniques to improve the training stability and convergence speed of the network.

[0021] In a preferred embodiment, the target text generation method based on big data is applied to the big data text generation scenario in the consulting, design and planning industry to process project reports, technical documents, market analysis and various types of document collections.

[0022] In a preferred embodiment, the target text generation system based on big data is used to execute the target text generation method based on big data, including:

[0023] Data acquisition module, used to obtain cutting-edge industry information from multiple data sources;

[0024] The structure prediction module is used to model the structure of the target document as a directed graph and use the graph attention network for analysis and prediction;

[0025] The association analysis module is used to construct the document association graph and document dependency graph to form a complete document knowledge network structure;

[0026] Content generation module, which generates text content based on graph structure information and graph propagation algorithm;

[0027] The dynamic update module is used to automatically trigger the graph propagation algorithm for intelligent update when the source data or related documents change.

[0028] The beneficial effects of the present invention are:

[0029] Through the intelligent structure prediction of graph neural networks, the limitations of traditional templated structures have been broken through, adaptive document organization based on content features has been achieved, and the logic and readability of generated documents have been significantly improved.

[0030] The construction and analysis technology of multi-document association graphs accurately identifies the complex association relationships between documents, forms an organic knowledge network, and improves the overall consistency and knowledge integration effect of the document set.

[0031] The application of graph propagation algorithm realizes the intelligent propagation of document updates, automatically maintains the dependencies and version consistency between documents, greatly reduces manual maintenance costs, and improves the system's automated management level. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the target text generation method based on big data of the present invention;

[0033] Figure 2 It is a histogram comparing the rationality scores of the document structures of the present invention;

[0034] Figure 3 is a line graph showing changes in the processing efficiency of the system of the present invention over time;

[0035] Figure 4 It is a radar chart comparing the performance of different algorithms of the present invention;

[0036] Figure 5 It is a pie chart of system resource consumption distribution of the present invention. DETAILED DESCRIPTION

[0037] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0038] At least one embodiment of the present invention discloses a method for generating target text based on big data, such as Figure 1 As shown, the following steps are included:

[0039] Step 1: Model the structure of the target document as a directed graph, and use the graph attention network to analyze the document structure graph. The attention algorithm calculates the importance weight of each node and predicts the optimal document structure organization method.

[0040] The specific steps include:

[0041] Step 1.1, model the structure of the target document as a directed graph:

[0042] ;

[0043] in, Represents a document structure diagram; A collection of nodes representing each part of a document, including hierarchical structures such as chapters, paragraphs, and subparagraphs; A set of edges representing hierarchical organizational relationships between parts.

[0044] Each node Represents a structural unit in a document, with a set of attributes including node type (such as chapter, section, paragraph, etc.), node level depth, node title, node content, etc.;

[0045] Each edge Represents the slave node To Node The hierarchical relationship shows yes The parent unit has directionality and can be assigned a weight value to represent the strength of the association.

[0046] This directed graph structure can fully express the organizational framework of the document and the subordinate relationships between its parts.

[0047] Step 1.2: Analyze the document structure graph using a graph attention network, calculate the importance weight of each node using an attention algorithm, and predict the optimal document structure organization method;

[0048] The calculation formula of the graph attention network is:

[0049] ;

[0050] in, It means that after the graph attention network calculation, Nodes in the layer The updated feature vector of , that is, the output feature of the current layer; Indicates the Neighbor nodes in the layer The feature vector of is used as the input feature of the current calculation; Represents the attention weight coefficient, which is used to measure the neighbor nodes For the current node The importance of the neighbor node. The higher the weight, the more important the information provided by the neighbor node. Representation node The set of neighbor nodes of Indicates the The trainable weight matrix of the layer is used to linearly transform node features; Indicates the node All neighbor nodes of Perform weighted sum operation; Represents nonlinear activation functions, such as ReLU and tanh, which are used to introduce nonlinear transformations to enhance the expressiveness of the model.

[0051] The graph attention network consists of a multi-head attention layer, a feature transformation layer, and an output layer.

[0052] The multi-head attention layer captures different types of document structure relationships by computing multiple attention heads in parallel;

[0053] The feature transformation layer performs nonlinear transformation on node features to enhance the expression ability;

[0054] The output layer generates the final document structure prediction results.

[0055] In consulting, design and planning scenarios, such as when generating a project feasibility report, the graph attention network can identify the strong correlation between the "market analysis" section and the "risk assessment" section, thereby adjusting the document structure to make the related content more logically close.

[0056] In some embodiments, the graph attention network adopts a multi-layer cascade structure, including 3 to 5 attention layers, and each layer uses 8 to 16 attention heads.

[0057] For example, for complex urban master planning documents, a 5-layer attention network with 16 attention heads in each layer can be used to fully capture the complex correlations between different planning elements.

[0058] In addition, residual connection and layer normalization techniques are used to improve the training stability and convergence speed of the network.

[0059] Step 2: Based on the predicted document structure, a document association graph is constructed. Graph neural networks are used to analyze the semantic similarity and logical association strength between documents to identify implicit relationships between documents.

[0060] The specific steps include:

[0061] Step 2.1, build document association graph:

[0062] ;

[0063] in, Represents the entire document association graph structure, which is a mathematical representation used to model the relationship between documents in a document collection; A collection of nodes representing each document in a document set; A set of edges representing the logical association relationship between documents.

[0064] In step 2.2, a graph neural network is used to analyze the semantic similarity and logical association strength between documents, identify the implicit relationship between documents, and generate a document association weight matrix to provide basic data for subsequent structural prediction.

[0065] In addition, the graph neural network includes a graph convolution layer, a pooling layer and a classification layer.

[0066] The graph convolution layer is responsible for feature propagation and aggregation on the document graph and learning the document representation vector;

[0067] The pooling layer aggregates graph-level features;

[0068] The classification layer outputs a score of the strength of association between documents.

[0069] In specific applications, such as when processing a document set for a large-scale infrastructure project, graph neural networks can automatically identify the strong dependency between the "environmental impact assessment report" and the "construction plan document" because the environmental impact assessment results directly affect the formulation of the construction plan.

[0070] Step 3: Based on the document association graph, a document dependency graph is constructed to clarify the dependency relationships and version correspondences between documents, thus forming a complete document knowledge network structure.

[0071] Build the document dependency graph:

[0072] ;

[0073] in, Represents a document dependency graph; Represents a collection of document nodes. Each node represents an independent document in the system and contains basic attribute information such as document ID, document type, and creation time. Represents a set of dependency edges between documents. Each edge defines the dependency type, dependency strength, and dependency direction between two documents. Represents the version information of a document, including the version history, version number, update timestamp, change summary, etc. of each document, which is used to track the evolution of the document.

[0074] Therefore, the model explicitly expresses the dependency and version correspondence between documents through the triple structure, providing a structured basis for the dynamic update of document collections.

[0075] The document dependency graph includes a dependency identification module, a version management module and a consistency checking module.

[0076] The dependency identification module determines the dependency strength between documents by analyzing document content and reference relationships;

[0077] The version management module maintains the version history and change records of each document;

[0078] The consistency check module verifies the logical consistency of a document set.

[0079] For example, in an urban planning project, when the "Transportation Planning Scheme" document is updated, the system can automatically identify that related documents such as "Land Use Planning" and "Environmental Protection Planning" need to be adjusted accordingly, and establish a clear dependency chain.

[0080] Step 4: Based on the constructed document structure graph, document association graph, and dependency graph, a graph propagation algorithm is used to generate text content.

[0081] The specific steps include:

[0082] In step 4.1, based on the constructed document association graph and dependency graph, the graph propagation algorithm is used to generate text content.

[0083] Among them, the graph propagation algorithm realizes information propagation through the following formula:

[0084] ;

[0085] in, Representation node Total propagation updates received, i.e., documents Updates that need to be made based on changes in related documents; Indicates the node The sum operation is performed on all neighboring nodes, that is, all other documents that are related to the current document are considered; Representation and Node The set of all adjacent nodes, that is, all other documents that are directly related to the current document i; Representation node With neighboring nodes The propagation weight between them ranges from 0 to 1. The larger the weight, the more documents Changes to the documentation The greater the influence, the weight is dynamically calculated based on factors such as the strength of the dependency relationship between documents, content relevance, and time relevance; Represents neighbor nodes Change information, including documents Changing factors such as content updates, structural adjustments, and data corrections are the input signals of the dissemination algorithm.

[0086] In step 4.2, the disseminated information is combined with the original big data content to generate target text content that meets the document structure requirements, thereby ensuring the relevance and consistency of the generated content.

[0087] The graph propagation algorithm includes an information aggregation unit, a weight calculation unit, and a content generation unit.

[0088] The information aggregation unit collects change information from related document nodes;

[0089] The weight calculation unit dynamically adjusts the propagation weight according to the correlation strength between documents;

[0090] The content generation unit generates new text content based on the aggregated information.

[0091] In practical applications, for example, when generating a corporate strategic planning report, after the data in the "Market Environment Analysis" section is updated, the graph propagation algorithm can propagate this change information to relevant chapters such as "Development Strategy Formulation" and "Risk Response Measures", ensuring that the content of the entire report remains consistent and timely.

[0092] In some implementations, the graph propagation algorithm uses a multi-round iterative propagation method and gradually converges to a stable state through 3-10 rounds of iterations.

[0093] Propagation weight Dynamic calculation is performed based on semantic similarity, citation frequency and time correlation between documents.

[0094] For example, when processing a set of technical documents for a new energy project, when key technical parameters in the "Technical Feasibility Analysis" document change, the algorithm first propagates the change information to the directly related "Equipment Selection Plan" document, then affects the "Investment Estimation Report" in the second round of propagation, and finally updates the "Project Risk Assessment" in the third round of propagation, thus achieving orderly information dissemination and intelligent content updating.

[0095] Step 5: When the source data or related documents change, the graph propagation algorithm is automatically triggered to intelligently update the structure and content of the affected documents based on the document dependencies.

[0096] In addition, through version identification and dependency tracking, the version consistency and logical integrity of the entire document set are maintained, and automated management of the document set is achieved.

[0097] In some implementations, dynamic updates employ an incremental update strategy, which updates only the affected portion of the document rather than regenerating the entire document, thereby improving update efficiency.

[0098] For example, when the population forecast data of a transportation planning project changes, the system only updates the traffic demand analysis chapter related to the population data, while keeping the content of other unrelated chapters unchanged, significantly reducing computing resource consumption and update time.

[0099] Application examples of this implementation:

[0100] Application scenarios:

[0101] A large-scale urban comprehensive development project required the generation of 120 related documents, including a master plan report, environmental impact assessment, transportation planning scheme, and economic feasibility analysis. These documents possessed complex logical connections and data dependencies, making it difficult to ensure consistency and timeliness among them using traditional methods.

[0102] Implementation process example:

[0103] First, the system models the master planning report as a document structure diagram containing 15 main chapters, and each chapter contains 3-8 sub-paragraph nodes.

[0104] By analyzing the semantic associations between chapters, the graph attention network identified a strong correlation weight of 0.85 between "land use planning" and "transportation planning", and a correlation weight of 0.72 between "environmental protection" and "industrial layout".

[0105] Based on these association analyses, the system automatically adjusts the document structure and organizes the highly relevant chapters logically and tightly.

[0106] The system constructed a multi-document association graph containing 120 document nodes.

[0107] The graph neural network identified a strong bidirectional dependency relationship between the “Environmental Impact Assessment Report” and the “Master Plan Report”, with a correlation strength of 0.91;

[0108] There is a one-way dependency relationship between "transportation planning scheme" and "land use planning document", with a correlation strength of 0.78.

[0109] Based on this, the system established a complete document dependency graph and clarified 896 dependency relationships between 120 documents.

[0110] When the project's population forecast data was adjusted from the original 800,000 to 950,000, the graph propagation algorithm was automatically activated.

[0111] In the first round of dissemination, the change information was disseminated to the directly related “housing needs analysis” and “public facilities configuration” documents;

[0112] In the second round of dissemination, the impact was extended to the “Traffic Demand Forecasting” and “Infrastructure Sizing” documents;

[0113] In the third round of dissemination, the "Investment Estimate" and "Implementation Timeline" documents were ultimately affected.

[0114] The entire dissemination process involved content updates for 32 documents, and the system automatically regenerated the relevant chapters.

[0115] Technical effect verification:

[0116] Through the application of this implementation method, the document management effect of the urban comprehensive development project is significantly improved.

[0117] The document structure rationality score increased from 6.2 points to 8.5 points, an increase of 37%.

[0118] The expert review team's satisfaction with the document's logical consistency increased from 65% to 92%, an increase of 42%.

[0119] In terms of system efficiency, the document update time has been shortened from an average of 72 hours to 18 hours, an efficiency improvement of 75%.

[0120] When core data changes, the automatic update coverage of related documents reaches 94%, while the update coverage of traditional methods is only 56%.

[0121] Manual maintenance costs were reduced by 68%, and the overall workload of project document management was reduced by approximately 60%.

[0122] Figures 2 to 5 They are comparison of document structure rationality scores; changes in system processing efficiency over time; performance comparison of different algorithms; and distribution of system resource consumption.

[0123] These data fully verify the significant advantages of the technical solution of this application in improving document quality and management efficiency, and provide effective technical support for the intelligent management of documents in large and complex projects.

[0124] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A target text generation method based on big data, characterized in that: The following steps are involved: Model the structure of the target document as a directed graph, where each node represents a structural unit in the document and has node type, hierarchical depth, node title, and node content; Each edge represents a hierarchical relationship from node i to node j, indicating that i is the parent unit of j, with directionality and association strength weight value. The graph attention network is used to analyze the document structure graph, and the importance weight of each node is calculated through the attention algorithm to predict the optimal document structure organization method. Based on the predicted document structure, a document association graph is constructed. Graph neural networks are used to analyze the semantic similarity and logical association strength between documents to identify implicit relationships between documents. Based on the document association graph, a document dependency graph is constructed. This includes a dependency identification module that analyzes document content and reference relationships to determine the dependency strength between documents. A version management module maintains the version history and change records of each document. A consistency check module verifies the logical consistency of the document set, clarifies the dependency relationships and version correspondences between documents, and forms a complete document knowledge network structure. Based on the constructed document structure graph, document association graph, and dependency graph, the graph propagation algorithm is used to generate text content; When the source data or related documents change, the graph propagation algorithm is automatically triggered to intelligently update the structure and content of the affected documents based on the document dependencies.

2. The target text generation method based on big data according to claim 1 is characterized in that: The graph attention network adopts a multi-layer cascade structure, including 3 to 5 attention layers, each layer uses 8 to 16 attention heads. The graph attention network includes a multi-head attention layer, a feature transformation layer and an output layer.

3. The target text generation method based on big data according to claim 1 is characterized in that: The graph neural network includes a graph convolution layer, a pooling layer, and a classification layer. The graph convolution layer is responsible for feature propagation and aggregation on the document graph, the pooling layer summarizes the graph-level features, and the classification layer outputs the association strength score between documents.

4. The method for generating target text based on big data according to claim 1, characterized in that: The graph propagation algorithm adopts a multi-round iterative propagation method and gradually converges to a stable state through 3 to 10 rounds of iterations. The graph propagation algorithm includes an information aggregation unit, a weight calculation unit and a content generation unit.

5. The method for generating target text based on big data according to claim 4, characterized in that: The propagation weight is dynamically calculated based on the semantic similarity, citation frequency and temporal correlation between documents.

6. The method for generating target text based on big data according to claim 1, characterized in that: Dynamic updates use an incremental update strategy, updating only the affected document parts.

7. The method for generating target text based on big data according to claim 2, characterized in that: The graph attention network uses residual connection and layer normalization techniques to improve the training stability and convergence speed of the network.

8. The method for generating target text based on big data according to claim 1, characterized in that: It is used in big data text generation scenarios in the consulting, design and planning industries to process project reports, technical documents, market analysis and various types of document collections.

9. A target text generation system based on big data, used to execute the target text generation method based on big data according to any one of claims 1 to 8, characterized in that: include: Data acquisition module, used to obtain cutting-edge industry information from multiple data sources; The structure prediction module is used to model the structure of the target document as a directed graph and use the graph attention network for analysis and prediction; The association analysis module is used to construct the document association graph and document dependency graph to form a complete document knowledge network structure; The content generation module generates text content based on graph structure information and graph propagation algorithm; the dynamic update module is used to automatically trigger the graph propagation algorithm for intelligent update when the source data or related documents change.

Images