AI-driven multilingual intelligent document abstract and mind map generation system

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an AI-driven multilingual intelligent document abstract and mind map generation system, which realizes unified representation of data such as multilingual texts, images and tables through a multi-modal semantic feature generation module, and constructs a dynamically optimized causal semantic map; generating a semantic flow path with clear logic based on a dynamic semantic flow modeling method, wherein the semantic flow path is used for multi-language abstract generation; the structure of the mind map is optimized by using the generative adversarial network, and the content of the map is further perfected through user interaction adjustment and logic consistency verification. The system supports a user to efficiently extract document core information, a logic structure is displayed in a visual mode, the accuracy, logicality and efficiency of document generation are remarkably improved, and the system is suitable for generation and decision support of complex multi-modal documents.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI-driven multilingual intelligent document summarization and mind map generation system. Background Art

[0002] With the popularity of big data and multilingual texts, users need to efficiently extract the core information of documents and present the logical relationships in a visual way to facilitate understanding and decision-making. However, traditional methods often have deficiencies in multimodal semantic integration and logical presentation, and cannot meet the processing needs of complex documents. The application of AI technology in document generation and semantic analysis provides new possibilities for improving document processing efficiency and accuracy.

[0003] The existing technology (Chinese invention patent, publication number: CN118551740B, name: document generation method, system and electronic device) uses a large language model to process document generation and mind map construction. Although the original document is sliced ​​using slicing technology and the target document and mind map are generated in combination with prompt information, there are the following major defects: the generated content is often too broad or inconsistent with the user's intention, especially when precise semantic extraction is required; the existing technology adjusts the target document by modifying the mind map feedback, but due to the single logic of mind map generation, it is easy to cause the adjusted content to still not fully meet user needs; the large language model needs to reprocess all original documents when regenerating the target document, which is a large workload and low efficiency. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides an AI-driven multilingual intelligent document summary and mind map generation system. The present invention uses a multimodal semantic feature generation module to uniformly represent data such as text, pictures, and tables, and construct dynamically optimized causal semantic graph data; the summary generation module uses a dynamic semantic flow modeling method to generate multilingual summaries; the mind map generation and optimization module optimizes the map structure based on the generative adversarial network, and further improves the system function through user interaction and logic verification. The system of the present invention can efficiently generate document summaries and mind maps with clear logic and accurate semantics, and realize the deep integration and intelligent presentation of multimodal data.

[0005] An AI-driven multilingual intelligent document summarization and mind map generation system, including: A multimodal semantic feature generation module is used to receive multimodal document data, extract and fuse semantic features of text content, image data, table data and video subtitle data, and generate unified multimodal semantic feature data; The semantic graph construction module constructs dynamically optimized causal semantic graph data representing the causal relationship between semantic nodes based on multimodal semantic feature data, including causal relationship extraction and multimodal relationship calibration, and generates calibrated causal relationship data; The summary generation module extracts key semantic node content based on the dynamic optimization of causal semantic graph data, uses the dynamic semantic flow modeling method to generate semantic flow paths, and combines the multilingual generation method to generate multilingual summary data; The mind map generation and optimization module generates the initial mind map based on the dynamic optimization of causal semantic graph data and multilingual summary data, and generates the final optimized mind map data through hierarchical relationship optimization and user interaction adjustment.

[0006] Preferably, in the multimodal semantic feature generation module: The text processing unit performs a word segmentation algorithm on the text content to decompose the semantic units, uses a grammar parsing method to establish a syntactic tree structure, and extracts the text topic through a semantic analysis model; The image processing unit extracts the region boundaries and semantic categories in the image through the image semantic segmentation method, and locates the coordinates and category information of each object in the image in combination with the object detection model; The table processing unit identifies the association between column headers, row headers and cell contents through a table parsing algorithm, and generates table hierarchical structure data; The video subtitle processing unit transcribes the video audio content into text through speech recognition technology, and uses the timing analysis method to mark the timestamp and semantic hierarchy of the subtitle text.

[0007] Preferably, the semantic graph construction module extracts the causal relationship between semantic nodes by the following method: Use the Bayesian network model to model the statistical correlation of multimodal semantic feature data and establish the potential causal relationship between nodes; The validity of causal relationships is verified using the conditional independence test method, the causal strength between nodes is calculated, and the edges with strengths below a predetermined threshold are removed; The causal relationships that have passed the conditional independence test are recorded as causal relationship data for further construction of the semantic graph.

[0008] Preferably, the multimodal relationship calibration in the semantic graph construction module includes: A cross-modal relationship graph network is used to optimize the semantic alignment between nodes of different modalities, where the feature representation of the nodes is calibrated by calculating the similarity of semantic vectors between modalities. Based on context consistency, the low-confidence edges in the graph are adjusted, the weights of high-confidence edges are increased, the weights of low-confidence edges are reduced, and invalid edges are removed; The calibrated nodes and edges are constructed as calibrated causal relationship data to dynamically optimize the further generation of causal semantic graphs.

[0009] Preferably, the summary generation module generates a semantic flow path by a dynamic semantic flow modeling method, comprising the following steps: Initialize the semantic flow matrix and calculate the semantic flow intensity according to the node weights in the causal semantic graph; The directionality of the semantic flow path is optimized based on the reinforcement learning method, where the target reward function for the semantic flow includes the path semantic coverage, the consistency of the flow direction, and the maximization of the node weight; Extract key node contents along the semantic flow path and generate semantic flow path data for multilingual summary generation.

[0010] Preferably, the multilingual summary data generated by the summary generation module is verified by a consistency verification method, comprising the following steps: A multilingual natural language inference model is used to verify the logical consistency of the abstracts in different language versions. The verification model uses the semantic similarity and logical coherence between sentences as evaluation indicators. Mark the inconsistent summary paragraphs, feed the inconsistent parts back to the summary generation module, and regenerate the optimized multilingual summary content.

[0011] Preferably, the mind map generation and optimization module generates the hierarchical structure of the mind map by the following method: Sort the semantic weights of map nodes and divide the priority levels of nodes according to the sorting results; According to the causal relationship between nodes, the lower-level nodes are associated with the higher-level nodes to form a hierarchical map structure; During the map generation process, ensure that the semantic association path between each node and its upper and lower nodes is complete and correct.

[0012] Preferably, the user interaction adjustment function of the mind map generation and optimization module includes: Provides a visual interface for the map, showing the semantic content of nodes and their causal relationship paths; Support users to adjust the position of nodes by dragging them, modify node content or add annotations through input interaction, and correct the adjusted node weights and edge relationships through real-time feedback; Automatically update the map structure modified by the user to the user's adjusted and optimized mind map data.

[0013] Preferably, the mind map generation and optimization module verifies the consistency between the mind map and the summary content through a natural language reasoning model, including the following steps: Use the inference model to verify whether the map node content matches the summary semantics, where the inference model uses the node semantic accuracy and summary matching rate as evaluation criteria; Verify whether the edge relationship in the map conforms to the logical chain in the summary content, and mark the nodes or edges that are inconsistent with the verification.

[0014] Preferably, the mind map generation and optimization module optimizes the mind map structure by generating an adversarial network, comprising the following steps: The generator generates multiple candidate graph structures, each of which contains redistributed node weights and edge relationships; The discriminator evaluates the logical consistency and verifies the semantic coherence of the candidate mind maps, and selects the optimal map structure as the final optimized mind map data.

[0015] Compared with the prior art, the advantages and beneficial effects of the present invention are: The present invention realizes the unified semantic representation of multilingual texts and non-text data such as images and tables by means of multimodal data fusion technology, thereby improving the accuracy and semantic consistency of generated content; The present invention uses dynamic semantic flow modeling and reinforcement learning optimization technology to achieve clear expression of summary logic and dynamic optimization of semantic flow paths, significantly improving the logic and coherence of summary generation; The present invention optimizes the mind map structure by generating adversarial networks, realizes visualization map optimization with strong logical consistency and clear hierarchy, and solves the problem of single logic presentation in comparison schemes. The present invention significantly improves the flexibility of the system and meets diverse user needs through user interactive adjustment and real-time optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural block diagram of the system of the present invention; Figure 2 It is a schematic diagram of the internal process of the semantic graph construction module in the present invention; Figure 3 It is a schematic diagram of the internal flow of the summary generation module in the present invention; Figure 4 It is a schematic diagram of the internal flow of the mind map generation and optimization module in the present invention. DETAILED DESCRIPTION

[0017] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure.

[0018] like Figure 1 As shown, an AI-driven multilingual intelligent document summarization and mind map generation system includes: A multimodal semantic feature generation module is used to receive multimodal document data, extract and fuse semantic features of text content, image data, table data and video subtitle data, and generate unified multimodal semantic feature data; Preferably, in the multimodal semantic feature generation module: The text processing unit performs a word segmentation algorithm on the text content to decompose the semantic units, uses a grammar parsing method to establish a syntactic tree structure, and extracts the text topic through a semantic analysis model; The image processing unit extracts the region boundaries and semantic categories in the image through the image semantic segmentation method, and locates the coordinates and category information of each object in the image in combination with the object detection model; The table processing unit identifies the association between column headers, row headers and cell contents through a table parsing algorithm, and generates table hierarchical structure data; The video subtitle processing unit transcribes the video audio content into text through speech recognition technology, and uses the timing analysis method to mark the timestamp and semantic hierarchy of the subtitle text.

[0019] The core goal of the multimodal semantic feature generation module is to extract and fuse multimodal semantic features from document data of different modalities to form a unified high-dimensional semantic representation, which provides a basis for subsequent semantic analysis, summary generation and mind map construction. Multimodal document data includes text content, image data, table data and video subtitle data. Different modalities have their own structures and characteristics, so it is necessary to use adaptive processing methods to extract their semantic features respectively, and use cross-modal fusion methods to uniformly map these features to a high-dimensional semantic space to generate unified multimodal semantic feature data.

[0020] The processing of text content is mainly achieved through word segmentation, grammatical parsing and semantic analysis models. The word segmentation algorithm is used to divide the text content into basic semantic units (such as words or phrases); then, the grammatical parsing method is used to generate a syntactic tree structure, which includes nodes (words or phrases) and their grammatical dependencies; finally, the main topic of the text is extracted through the semantic analysis model.

[0021] For example, when processing a text describing the weather, the word segmentation algorithm divides it into word units such as "today", "weather", and "sunny"; the grammatical parsing generates syntactic trees of "today-subject-weather", "weather-modifier-sunny"; and the semantic analysis model recognizes "sunny weather" as a topic feature.

[0022] Image data processing is completed through a combination of image semantic segmentation and object detection. The image semantic segmentation method is used to segment different regions in the image, extract region boundaries and semantic categories; the object detection model further identifies the specific category of each object in the image (such as cars, buildings) and its coordinate location.

[0023] For example, when processing a city landscape picture, the image semantic segmentation method divides the picture into areas such as "sky", "road", and "building"; the object detection model identifies specific objects in the "building" area such as "high-rise buildings" or "bridges" and records the coordinates and categories of these objects.

[0024] The processing of table data is completed by the table parsing algorithm. The algorithm first identifies the column headers and row headers, determines the relationship between the cell content and the header through association analysis, and then generates hierarchical table structure data based on the cell content and its position in the table.

[0025] For example, for a sales table, the algorithm recognizes "Product Name" as the column header and "2024" as the row header. The data "1000" in the cell is parsed as "the sales volume of a certain product in 2024 is 1000" through position association. The generated hierarchical structure associates the title and cell content into complete semantic information.

[0026] The processing of video subtitle data is completed by combining speech recognition and time series analysis. The speech recognition model transcribes the audio content in the video into text and records the timestamp corresponding to each text segment; the time series analysis method further extracts the contextual semantic relationship and time hierarchy of the subtitle text to generate structured subtitle semantic features.

[0027] For example, when processing a speech video, the speech recognition model transcribes the audio into multiple sentences of subtitle text, and marks "0:00-0:10" for the first sentence and "0:10-0:20" for the second sentence; the timing analysis method identifies the logical relationship between these sentences (such as cause and effect or progression) and generates subtitle features that associate time with semantics.

[0028] The semantic features of the above modal data are mapped to a unified high-dimensional semantic space. The cross-modal fusion method is implemented through a cross-modal transformation network (CMTN), which receives multimodal semantic features as input and generates a unified semantic vector representation. Specifically, each modal feature vector is converted into a high-dimensional semantic vector through a modality-specific encoder, and then the semantic differences between modalities are aligned through an attention mechanism in the fusion layer to generate unified multimodal semantic feature data.

[0029] The semantic alignment of different modal data is achieved, and the generated unified multimodal semantic feature data has high-dimensional semantic representation capabilities, providing a basis for subsequent modules to be compatible with multimodal data. Through technical means such as word segmentation, semantic segmentation, object detection, and time series analysis, the core semantic information of text, pictures, tables, and subtitles can be accurately extracted to avoid information loss. The use of cross-modal transformation networks and efficient pre-trained models significantly reduces the complexity of processing multimodal data and improves the efficiency of semantic feature extraction and fusion.

[0030] In the embodiment, a user uploads a business report containing text description, sales data table, product pictures and video explanation. The goal is to extract and unify the semantic features of these modalities to generate a business decision summary and mind map. The processing steps include: The text processing unit breaks down the descriptive text in the report into words through a word segmentation algorithm, and extracts “quarterly sales performance” and “product market feedback” as topics through a semantic analysis model.

[0031] The image processing unit performs semantic segmentation and object detection on product images, and extracts the semantic features of "product appearance", "product performance display" and their corresponding area locations.

[0032] The table processing unit parses the sales table data and generates a hierarchical semantic structure such as "the sales volume of product A in 2024 is 1,000 units".

[0033] The video subtitle processing unit transcribes the audio content of the commentary video into text subtitles, marks the timestamp, and extracts the logical relationship between the subtitle texts (such as reason explanation).

[0034] The cross-modal fusion method inputs the feature vectors of text, pictures, tables and video subtitles into the cross-modal transformation network, aligns the semantic differences through the attention mechanism, and generates a unified semantic vector representation.

[0035] The generated multimodal semantic feature data covers text themes, image object features, table hierarchical data, and subtitle logical information. The unified high-dimensional semantic vector can support the generation of clear business decision summaries and logically rigorous mind maps.

[0036] like Figure 2 As shown, the semantic graph construction module constructs dynamic optimized causal semantic graph data representing the causal relationship between semantic nodes based on multimodal semantic feature data, including causal relationship extraction and multimodal relationship calibration, and generates calibrated causal relationship data; The semantic graph construction module aims to construct a dynamically optimized causal semantic graph based on multimodal semantic feature data to represent the causal relationship between semantic nodes. This module mainly consists of two key parts: causal relationship extraction and multimodal relationship calibration. By extracting semantic nodes and their causal relationships from multimodal data and performing multimodal calibration optimization on them, a causal semantic graph with dynamic update capabilities can be generated. This graph can not only accurately describe the logical relationship between semantic nodes, but also adapt to the dynamic changes of multimodal features, ensuring efficient collaboration with subsequent modules (such as summary generation and mind map optimization).

[0037] Based on multimodal semantic feature data, causal reasoning methods are used to extract causal relationships between semantic nodes. Causal reasoning models the associations between nodes through Bayesian networks and generates a preliminary structure of causal relationships. Specifically, the feature vector of each semantic node is mapped to a node in the network, and the edges between nodes represent possible causal relationships. The reliability of the edges is further verified using conditional independence tests, and the strength of the causal relationship is calculated.

[0038] Multimodal relation calibration aims to optimize the semantic consistency of nodes in different modalities in the initial causal relationship graph. Through the Cross-modal Relational Graph Network (CR-GNN), the alignment and consistency enhancement of node semantic representation are achieved. The calibration process includes the following steps: Cross-modal alignment: Map the feature vectors of nodes in different modalities to the same semantic space, and adjust the weights of the nodes by calculating the cosine similarity between the modalities, so that the cross-modal node representation has a high consistency.

[0039] Relationship optimization: adjust the edges with low confidence in the initial causal relationship graph, remove incorrect causal relationship edges through context consistency verification, and enhance the weights of high confidence edges.

[0040] Through the above two core steps, the generated calibrated causal relationship data can accurately and comprehensively describe the causal relationship between semantic nodes while ensuring that it is consistent with the characteristics of multimodal data.

[0041] Preferably, the semantic graph construction module extracts the causal relationship between semantic nodes by the following method: Use the Bayesian network model to model the statistical correlation of multimodal semantic feature data and establish the potential causal relationship between nodes; The validity of causal relationships is verified using the conditional independence test method, the causal strength between nodes is calculated, and the edges with strengths below a predetermined threshold are removed; The causal relationships that have passed the conditional independence test are recorded as causal relationship data for further construction of the semantic graph.

[0042] The causal relationship extraction function in the semantic graph construction module aims to accurately identify the causal relationship between semantic nodes based on multimodal semantic feature data to support the generation of dynamic optimization semantic graphs. The process uses the Bayesian network model to model the statistical correlation between semantic nodes and construct potential causal relationships; then, the validity of the potential relationship is verified by the conditional independence test method, and the strength of the causal relationship is calculated; finally, the tested high-confidence causal relationship is recorded as causal relationship data for further optimization and application of the graph.

[0043] In the present invention, the Bayesian network is used to establish the preliminary causal relationship structure of multimodal semantic feature data. Each semantic node represents a random variable, and its feature vector is provided by the semantic feature generation module (such as the topic feature vector of the text node and the object feature vector of the picture node). The Bayesian network constructs possible causal relationship edges by calculating the joint probability distribution between nodes.

[0044] For example, in the user review semantic analysis scenario, the node "stable performance" and the node "increased sales" may have statistical correlation. Through Bayesian network modeling, it can be inferred that "stable performance" is the potential cause of "increased sales".

[0045] Potential causal relationships need to be verified through conditional independence tests. The test process analyzes whether there is still a statistical correlation between other nodes when the conditions of a node are known, and then determines the authenticity of the causal relationship. The conditional independence test method used is based on calculating the causal strength between nodes: , in, Representation Node For Node the causal strength of Indicates that only the condition is a node When the node The probability of Indicates that the condition is a node and context node When the node The probability of When it is less than a predetermined threshold, the corresponding edges are removed to reduce invalid relationships.

[0046] For high-confidence causal relationships verified by the conditional independence test, they are recorded as causal relationship data in the form of node-edge-weight. Nodes represent semantic features, edges represent causal relationships, and weights represent causal strengths. The generated causal relationship data serves as an important basis for semantic graph optimization and summary generation.

[0047] Through dual screening of Bayesian network modeling and conditional independence test, we ensure that only causal relationships with high confidence are recorded, reducing misleading or noisy relationships. The extraction of causal relationships provides global semantic association information for the semantic map, which is convenient for subsequent modules to efficiently use in summary generation and mind map construction. By removing low-intensity edges and readjusting weights, the generated causal relationship data can adapt to the dynamic changes of multimodal semantic features.

[0048] In this example, a user uploads a multimodal document containing product review text, pictures, and sales statistics. The goal is to extract the causal relationship between product characteristics and sales changes to generate data-driven marketing decisions. Processing steps: (1) Bayesian network modeling: The “stable performance” and “appearance design” extracted from the comments are used as semantic nodes. The object features in the accompanying pictures (such as “highlight product appearance”) and the “sales increase” node in the sales data are combined to construct a preliminary causal network. Possible causal edges are generated by calculating the joint probability distribution between nodes: such as “stable performance → sales increase” and “appearance design → user satisfaction”.

[0049] (2) Conditional independence test: Perform a conditional independence test on each edge in the preliminary causal network and calculate the causal strength For the edge “Design → User Satisfaction”, if = 0.4 and is lower than a predetermined threshold (such as 0.5), then the edge is removed; for the edge “stable performance → increased sales”, if =0.8 is higher than the threshold, then it is retained and weighted.

[0050] (3) Generate causal relationship data: record the node and edge weights of the final causal network. The generated causal relationship data include "stable performance → increased sales (weight 0.8)" and "user experience → user satisfaction (weight 0.7)", which provide support for subsequent semantic graph optimization and decision summary generation.

[0051] Preferably, the multimodal relationship calibration in the semantic graph construction module includes: A cross-modal relationship graph network is used to optimize the semantic alignment between nodes of different modalities, where the feature representation of the nodes is calibrated by calculating the similarity of semantic vectors between modalities. Based on context consistency, the low-confidence edges in the graph are adjusted, the weights of high-confidence edges are increased, the weights of low-confidence edges are reduced, and invalid edges are removed; The calibrated nodes and edges are constructed as calibrated causal relationship data to dynamically optimize the further generation of causal semantic graphs.

[0052] The multimodal relationship calibration of the semantic graph building module aims to optimize the feature representation and causal credibility between semantic nodes of different modalities to improve the overall consistency and accuracy of the semantic graph. The calibration method optimizes the alignment of semantic vectors of nodes of different modalities through a cross-modal relationship graph network, and adjusts low-confidence causal edges based on contextual consistency, thereby eliminating invalid edges and strengthening credible relationships.

[0053] The semantic features of multimodal data come from different modalities, such as text, images, tables, and video subtitles, and their semantic expressions have significant modal differences. The Cross-modal Relational Graph Network (CR-GNN) constructs a relational graph of multimodal nodes and maps the features of nodes in different modalities into a unified high-dimensional semantic space, thereby achieving semantic alignment.

[0054] The semantic feature vector of each modal node is expressed as , whose initial value is provided by the semantic feature generation module. The edges between nodes represent possible semantic associations between modalities, and their weights are obtained by calculating the cosine similarity of the node feature vectors: , in, Representation Node and The semantic similarity of is the Euclidean norm of the vector.

[0055] Through a multi-layer message passing mechanism, CR-GNN shares contextual information among multimodal nodes and updates the feature vector representation of the nodes, gradually aligning the semantic features of different modalities.

[0056] After optimizing the node semantic alignment, the edges with lower confidence in the causal graph are further calibrated. The context consistency calibration includes the following steps: Each edge is re-evaluated based on the consistency of the context, and the edge weight with high confidence is is promoted, and the weights of low-confidence edges are reduced. Context consistency is calculated by the following formula: , in, Represents edge The contextual consistency score of For edge The associated context edge set. When the calibrated weight of an edge is lower than a preset threshold (such as 0.3), it is directly removed to reduce noise.

[0057] After calibration is completed, the optimized nodes and their edge relationships are recorded as calibrated causal relationship data in the structure of node-edge-weight. The calibrated causal relationship data contains the optimized semantic node representation, edge weight and its causal strength, which is used to dynamically optimize the causal semantic graph.

[0058] The cross-modal relationship graph network realizes the alignment of semantic features of different modalities in a unified space, so that multimodal nodes have consistent semantic expression capabilities. Context consistency calibration strengthens the edge weights of high confidence, removes invalid edges of low confidence, and improves the reliability of causal relationships in the semantic graph. The optimized calibrated causal relationship data provides high-quality input for dynamically optimizing the causal semantic graph, which can quickly adapt when multimodal data changes.

[0059] In this example, a user submits a set of product review data, including text reviews, product images, and sales table data. The goal is to analyze the causal relationship between characteristics such as "stable performance" and "appearance design" and "sales increase" through semantic graph construction. Processing steps: The “Design” node in the comment text generates a feature vector through semantic analysis . Object feature vectors for the “highlight appearance” region in the product image Generated through object detection. Feature vector of the "sales increase" node in the table data Generated by table parsing. CR-GNN updates these node feature vectors through message passing and aligns them in the semantic space, making the semantic correlation between "appearance design" and "sales increase" higher.

[0060] In the initial causal relationship graph, the edge weight of "Appearance Design → User Satisfaction" is 0.4. After context consistency calibration, its weight is reduced to 0.2 due to insufficient support for related edges, which is lower than the threshold of 0.3 and is therefore removed. The edge weight of "Stable Performance → Sales Increase" is increased from 0.7 to 0.85, and is ultimately retained as a high-confidence edge due to the strong consistency of its contextual support (such as "user experience").

[0061] The calibrated causal relationship data finally generated records the high-confidence causal relationship of "stable performance → increased sales (weight 0.85)" and eliminates low-confidence edges such as "appearance design → user satisfaction", providing optimized input for the causal semantic graph.

[0062] The calibrated semantic graph accurately reflects the key causal relationships in multimodal data, clearly reveals how product characteristics affect user experience and sales changes, and provides a reliable basis for subsequent marketing strategy decisions.

[0063] like Figure 3As shown, the summary generation module extracts key semantic node contents based on the dynamic optimization of causal semantic graph data, uses the dynamic semantic flow modeling method to generate semantic flow paths, and combines the multilingual generation method to generate multilingual summary data; The core of the summary generation module is to extract the content of key semantic nodes based on dynamic optimization of causal semantic graph data, generate semantic flow paths through dynamic semantic flow modeling, and combine multilingual generation methods to generate accurate and coherent multilingual summary data. The implementation of this module involves three key parts: semantic node extraction, semantic flow modeling, and multilingual generation.

[0064] Dynamically optimized causal semantic graph data is a high-quality structured representation of semantic nodes and their causal relationships. The summary generation module first extracts key semantic nodes based on node weights and causal strength. The calculation of node weights is based on semantic importance and centrality indicators in the causal network, such as using the node's PageRank value to measure its importance. During the extraction process, high-weight nodes are given priority, while ensuring that the extracted nodes are logically coherent, that is, maintaining the integrity of the causal chain.

[0065] After the key nodes are extracted, the dynamic semantic flow modeling method is used to generate the semantic flow path and clarify the flow direction and logical sequence of the semantic nodes. The semantic flow modeling constructs the semantic flow matrix through the path optimization technology in the fluid mechanics model, and further optimizes the semantic flow path through the reinforcement learning algorithm to ensure that the semantic flow direction is consistent with the summary generation goals, such as summary length limit and topic focus.

[0066] The multilingual generation method is implemented based on a multilingual pre-trained generation model (such as mT5 or mBART). The extracted semantic flow path is used as the input of the model, and the multilingual summary content is generated in combination with the target language and style requirements. The model uses a semantic alignment mechanism during the generation process to ensure the consistency of the semantic expression and logical order of the summaries generated in different languages. In order to improve the generation quality, a consistency verification mechanism is used to check the logical consistency between multilingual versions and optimize the generation results.

[0067] Preferably, the summary generation module generates a semantic flow path by a dynamic semantic flow modeling method, comprising the following steps: Initialize the semantic flow matrix and calculate the semantic flow intensity according to the node weights in the causal semantic graph; The directionality of the semantic flow path is optimized based on the reinforcement learning method, where the target reward function for the semantic flow includes the path semantic coverage, the consistency of the flow direction, and the maximization of the node weight; Extract key node contents along the semantic flow path and generate semantic flow path data for multilingual summary generation.

[0068] The summary generation module generates semantic flow paths through dynamic semantic flow modeling to construct summary content with clear logic and compact semantics. The core of dynamic semantic flow modeling is to initialize the semantic flow matrix based on the node weights in the causal semantic graph, optimize the path directionality through reinforcement learning, and extract key node content to generate semantic flow path data, providing high-quality input for multilingual summary generation. The whole process covers three key steps: initializing the semantic flow matrix, optimizing the semantic flow path, and extracting key node content along the path.

[0069] The semantic flow matrix is ​​initialized based on the node weights and edge weights in the causal semantic graph. The semantic flow matrix is ​​used to describe the semantic flow intensity between nodes. Representation Node The importance of nodes in the causal graph is usually provided by the semantic feature generation module and calculated in combination with the semantic centrality index of the node (such as PageRank value). To Node Semantic flow strength Calculated by the following formula: , in, Representation Node arrive the strength of the causal relationship; Representation Node The weight of Representation Node The flow intensity matrix is ​​generated by calculating the flow intensity of all node pairs one by one, laying the foundation for subsequent path optimization.

[0070] The reinforcement learning method is used to optimize the semantic flow path to ensure that the logical order of the semantic nodes is consistent with the summary generation goal. The state, action, and reward functions of the reinforcement learning model are defined as follows: State: the node sequence of the current semantic flow path; Action: Select the next node to join the path; Reward function: Comprehensively consider the path semantic coverage, flow direction consistency and node weight maximization: , in, It represents the path semantic coverage and the total semantic weight of the selected node; Indicates the consistency of flow direction, which is calculated based on the coherence of flow intensity between nodes; Represents the sum of the weights of the nodes in the path; , , It represents the balance coefficient, which is used to adjust the relative importance of each indicator. The optimization process gradually adjusts the node selection strategy of the path through the policy gradient method, and finally generates a semantic flow path that meets the target requirements.

[0071] The optimized semantic flow path is used to extract the core content of each node in the path and generate semantic flow path data. The content of each node is represented by its semantic feature vector, which is directly obtained through the node representation of the causal semantic graph. The extracted semantic flow path data has logical coherence and semantic coverage, providing accurate input for multilingual summary generation.

[0072] Dynamic semantic flow modeling optimizes the path direction through reinforcement learning, ensures the logical order and semantic coherence of the summary content, and improves the readability of the summary. The key node content extracted based on the semantic flow path has high semantic importance and information coverage, ensuring that the summary can convey the core information of the document. The initialization and optimization process of the flow matrix can adapt to document content with different semantic structures and support the summary generation needs of multimodal data.

[0073] In this embodiment, a market analysis report is processed, which includes a sales data table, a competitive analysis text, and a market trend picture. The goal is to generate a multilingual summary covering the core ideas of market performance. Processing steps: Initialize the semantic flow matrix: extract nodes in the semantic graph, including "sales growth", "market share change" and "competition analysis". The node weights are , , The flow intensity matrix is ​​calculated as: , A similar method is used to calculate the flow intensity of all node pairs and generate a flow matrix.

[0074] Optimize the semantic flow path: The initialization path starts from "sales growth". Reinforcement learning optimizes the path direction according to the reward function and selects the path "sales growth → market share change → competition analysis". Reward function calculation: Path semantic coverage ; Flow direction consistency ; Total node weight .

[0075] Extract key node content: extract node content along the optimization path, "sales growth": the sales data table shows a growth rate of 15%; "market share change": the market trend chart shows a change in share from 20% to 25%; "competition analysis": the competition analysis text mentions the impact of a competitor's withdrawal from the market. The generated semantic flow path data describes the market growth logic chain.

[0076] The generated multilingual summary content expresses the logical main line of market performance: sales growth drives market share increase, and competition analysis further explains the reasons for share change. The English summary is: "Sales growth of 15% led to a market share increase from 20% to 25%, supported by competitive market dynamics." The Chinese summary is: "Sales growth of 15% led to a market share increase from 20% to 25%, supported by competitive market dynamics." Preferably, the multilingual summary data generated by the summary generation module is verified by a consistency verification method, comprising the following steps: A multilingual natural language inference model is used to verify the logical consistency of the abstracts in different language versions. The verification model uses the semantic similarity and logical coherence between sentences as evaluation indicators. Mark the inconsistent summary paragraphs, feed the inconsistent parts back to the summary generation module, and regenerate the optimized multilingual summary content.

[0077] The goal of the multilingual summary consistency verification method is to verify the logical consistency and semantic similarity of the summary content in different language versions, and to ensure that the generated multilingual summary is consistent in terms of expression logic, content consistency, and information coverage. This method is implemented through the Multilingual Natural Language Inference (M-NLI) model, which evaluates the semantic similarity and logical coherence between sentences in the multilingual version of the summary; the content of the annotated inconsistent parts is fed back to the summary generation module to regenerate the optimized multilingual summary content.

[0078] The Multilingual Natural Language Inference model (M-NLI) is based on multilingual semantic embedding technology and can verify the logical consistency of summary versions in different languages. The specific verification process is as follows: Generate a semantic embedding vector for each summary sentence through a multilingual embedding model (such as XLM-R) , used to represent the semantic features of a sentence. The semantic similarity of corresponding sentences in different language versions is calculated using the following formula: , in, Represents the source language sentence and target language sentences The semantic similarity of and Separate sentences and The semantic embedding vector of is the Euclidean norm of the vector. The inference model is used to check whether the logical relationships (such as causality, progression, and contrast) between sentences are consistent, for example, to verify the coherence of the "causal chain".

[0079] When the semantic similarity between sentences is lower than the preset threshold (such as 0.8) or the logical coherence check fails, the corresponding summary paragraph is marked as "inconsistent". The marked inconsistent parts are passed to the summary generation module through the feedback mechanism, and the optimized content is regenerated to improve consistency. The feedback optimization process combines the reinforcement learning model to adjust the summary generation parameters (such as node selection strategy or path weight distribution) to ensure that the optimized summary meets the consistency requirements.

[0080] Through semantic embedding and semantic similarity verification, we ensure that the multilingual summary content has the same core information and semantic expression in different languages. Logical verification verifies the causal chain and logical order between sentences through the reasoning model to avoid logical inconsistency caused by information bias in the translation or generation process. The annotation and feedback mechanism efficiently repairs inconsistent content, realizing dynamic interaction and optimization closed loop between the summary generation module and the verification module.

[0081] In this example, the system generates a multilingual summary of a market analysis report. The goal is to verify the logical and semantic consistency of the Chinese and English summaries through a consistency verification method to ensure that the multilingual summary can accurately express the key information of market performance. Processing steps: The Chinese summary sentence "15% sales growth leads to increased market share" generates a semantic embedding vector ; Generate semantic embedding vector for the English summary sentence “Sales increased by 15%, driving market share growth” .

[0082] By formula The similarity is calculated and the result is 0.92, which is higher than the threshold of 0.8 and is judged to be semantically consistent.

[0083] Verify whether the summary sentence "market share increase" is caused by "sales growth". The reasoning model confirms that the logical chain is correct through semantic analysis and the verification is passed. If the logic verification fails (for example, "the causal chain is missing"), it is marked as "inconsistent" and fed back to the summary generation module.

[0084] For paragraphs that failed logical verification, the system marked and regenerated the content. Assuming that the original sentence of the English abstract "Marketshare grew independently" does not match the logic, the system generates an optimized new sentence "Market sharegrowth was driven by sales increase".

[0085] The consistency check method successfully verified and optimized the multilingual summary content, ensuring the consistency of the Chinese and English summaries in logical order and semantic expression. The final summary example is as follows: Chinese summary: Sales increased by 15%, driving market share growth and significantly improving market performance. English summary: Sales increased by 15%, driving market share growth and significantly improving market performance. like Figure 4 As shown, the mind map generation and optimization module generates an initial mind map based on dynamically optimized causal semantic graph data and multilingual summary data, and generates the final optimized mind map data through hierarchical relationship optimization and user interaction adjustment.

[0086] The goal of the mind map generation and optimization module is to generate a logically clear and hierarchical initial mind map based on the dynamic optimization of the causal semantic map data and multilingual summary data, and to generate the final optimized mind map data through hierarchical relationship optimization and user interaction adjustment. This module combines the causal relationship of the nodes in the causal semantic map, the semantic information of the multilingual summary, and the real-time input of the user to gradually improve the structure of the mind map to ensure its logic and user-friendliness.

[0087] The initial mind map generation is based on the dynamic optimization of causal semantic graph data, and the initial structure of the mind map is constructed through the recursive graph neural network (R-GNN): the nodes in the semantic graph serve as the basic nodes of the map, and each node represents a semantic concept (such as "stable performance" or "sales growth"), and its attributes include semantic weight, modal source, and summary semantics. The causal relationship in the semantic graph is mapped to the edge in the map, and each edge contains weight and directionality, which is used to represent the causal or logical relationship between nodes. The initial mind map is structurally presented as a hierarchical directed graph, reflecting the core logic and semantic chain of the document content.

[0088] Hierarchical relationship optimization Based on the importance ranking of nodes and logical analysis of semantic relationships, the initial mind map is restructured: using node weights Nodes are ranked by importance, and nodes with higher weights are promoted to higher levels first. The weight calculation formula is: , in, Represents the semantic weight of the node, provided by the semantic graph; Indicates the centrality index of the node, such as PageRank value; , Represents the weight adjustment parameter, which is used to balance the impact of the two indicators.

[0089] According to the weight sorting of nodes, nodes are divided into different levels, with high-weight nodes located in higher levels and low-weight nodes allocated to sub-levels, to ensure the clear hierarchy of the mind map. Adjust the cross-level edge relationship to reduce the complexity of the cross-level path and ensure the simplicity of the mind map logic.

[0090] Support users to adjust the structure of mind maps through the interactive interface to enhance the customization and user-friendliness of mind maps: Provide dynamic visualization of nodes and edges, so that users can clearly view the hierarchical structure and logical relationship of the mind map. Allow users to drag nodes to adjust their positions, modify node content, or add comments. Support users to adjust edge weights or delete invalid edges, and update the mind map logic in real time. Based on user operations, the system adjusts the causal relationship and hierarchical division between nodes through an interactive real-time correction algorithm to ensure the consistency and accuracy of the mind map logic.

[0091] After user interaction adjustment, the mind map structure is further optimized through the Generative Adversarial Network (GAN): Generator: Generate multiple candidate mind maps to explore possible node layouts and edge relationships. Discriminator: Evaluate candidate mind maps based on logical consistency and semantic coherence, and select the optimal structure as the final optimized mind map data.

[0092] Preferably, the mind map generation and optimization module generates the hierarchical structure of the mind map by the following method: Sort the semantic weights of map nodes and divide the priority levels of nodes according to the sorting results; According to the causal relationship between nodes, the lower-level nodes are associated with the higher-level nodes to form a hierarchical map structure; During the map generation process, ensure that the semantic association path between each node and its upper and lower nodes is complete and correct.

[0093] The hierarchical structure generation method of the mind map generation and optimization module aims to divide the nodes in the dynamic optimization causal semantic map into layers according to semantic weights and causal relationships, and build a hierarchical and logically clear mind map structure. In the specific implementation process, the semantic weights of the nodes are first sorted, and the priority levels of the nodes are divided according to the sorting results; then, the low-level nodes are associated with the high-level nodes according to the causal relationship between the nodes, and it is ensured that the semantic association path between each node and its upper and lower nodes is complete and correct.

[0094] The node semantic weight reflects the importance of the node in the entire map. Its calculation combines the semantic feature weight of the node and the network characteristics in the map (such as centrality index). The specific steps are as follows: Node Weight It is obtained by weighted summation of semantic feature weight and centrality index, and the formula is: , in, The semantic feature weight of the node is provided by the dynamically optimized causal semantic graph; Centrality indicators of nodes (such as PageRank values); , The weight adjustment coefficient is used to balance the influence of semantic features and centrality indicators. All nodes are sorted in descending order according to the calculated weights, and high-weight nodes are assigned to higher levels, and nodes with lower weights are assigned to corresponding sub-levels. The hierarchical division ensures that high-weight nodes form the backbone structure, and low-weight nodes form sub-levels around high-weight nodes.

[0095] Based on the causal relationship in the semantic graph, the low-level nodes are associated with the high-level nodes through edge relationships to build a hierarchical map structure: the weight of each edge Representation Node and nodes The strength of the causal relationship between nodes. Located on a high level, and If the value is higher than the preset threshold, the node Associated to a Node If the edge weight is low or does not conform to the causal logic, the corresponding edge is removed to reduce invalid associations.

[0096] To ensure that the semantic association path between each node and its upper and lower nodes is complete, the map generation process needs to verify whether all causal chains between the node and its upper and lower nodes are connected. For interrupted chains, the module will use the path repair algorithm to supplement the missing edges or adjust the node position to make the logical chain complete.

[0097] Through semantic weight sorting and causal relationship driven association, the generated mind map has a clear structure and highlights the core semantic nodes and logical main line in the document. Path integrity verification ensures that the causal chain between nodes is accurately presented in the mind map, avoiding information breaks or unclear logic. Hierarchical division and path optimization improve the readability and ease of use of the mind map, allowing users to quickly understand the core content and logical framework of the document.

[0098] In this example, a product report is processed, which includes multimodal semantic information such as "sales growth", "increased user satisfaction", and "increased market share". The goal is to generate a clear mind map to show product performance and its driving factors. Processing steps: The extracted semantic nodes include "Sales Growth", "User Satisfaction" and "Market Share". Calculate the node weights: (because it appears most frequently in the report and has the most associated nodes); ; . “Sales growth” is classified as the highest level, and “market share increase” and “user satisfaction improvement” are classified as secondary levels.

[0099] Causality-driven hierarchical associations determine edge weights: The edge weight of "increased user satisfaction → increased sales" is , keep this edge; the edge weight of “market share increase → sales growth” is , keep this edge. Associate "improved user satisfaction" and "increased market share" with "sales growth" to build a backbone structure with "sales growth" as the core.

[0100] Path integrity assurance, check chain integrity: verify whether the path of "increased user satisfaction → increased sales → increased market share" is connected. Supplement missing edges. If it is found that the logic between "increased market share" and "increased user satisfaction" is not connected, add auxiliary edges through the path repair algorithm to ensure the integrity of the chain.

[0101] The generated mind map shows the driving factors of "sales growth", including the logical relationship between "improved user satisfaction" and "increased market share". The map has clear levels and logical coherence, providing intuitive support for marketing decisions.

[0102] Preferably, the user interaction adjustment function of the mind map generation and optimization module includes: Provides a visual interface for the map, showing the semantic content of nodes and their causal relationship paths; Support users to adjust the position of nodes by dragging them, modify node content or add annotations through input interaction, and correct the adjusted node weights and edge relationships through real-time feedback; Automatically update the map structure modified by the user to the user's adjusted and optimized mind map data.

[0103] The user interaction adjustment function of the mind map generation and optimization module is designed to provide users with the ability to visually adjust the map structure through an interactive interface, ensuring that the map is more in line with user needs in terms of logical integrity and content customization. Users can adjust the position of nodes by dragging them, modify node content or add comments. At the same time, the system automatically updates the weights and edge relationships between nodes based on the user's adjustments and generates an optimized map structure. The entire interactive process includes three key steps: visual interface presentation, real-time adjustment and correction, and dynamic update optimization.

[0104] The visualization interface is the core window for users to operate the map, showing the semantic content of the node and its causal relationship path. Node display: Each node displays its semantic content and weight. The node color or size is dynamically adjusted according to the weight. The higher the weight, the more prominent the node. Edge relationship display: The edge between nodes is represented by an arrow to indicate the direction of the causal relationship, and the thickness of the edge indicates the weight strength. High-weight edges are thicker, and low-weight edges are thinner or dotted. Real-time response: When the user clicks or hovers a node, the interface highlights all directly related nodes and paths of the node, making it easier for users to understand the contextual relationship of the node.

[0105] Users can modify the map through interactive operations, and the system will correct the weights and edge relationships between nodes in real time to maintain logical consistency. Users can adjust the position of nodes in the map by dragging them, and the adjusted position will trigger the redistribution of weights between nodes. The weight adjustment formula is: , in, Represents the adjusted node and nodes The weight of , Represents the adjusted node and The location coordinates of Represents the adjusted node and The user can directly input and modify the semantic content of the node or add comments to the node. The comment content is distinguished by color or special marks. The user can manually delete invalid edges, or enter the weight of the specified edge, and the system will update the edge relationship in real time.

[0106] After the user's adjustment is completed, the system dynamically updates and optimizes the mind map: deletes isolated nodes or invalid edges to maintain the logical integrity of the cause-effect relationship; recalculates node weights based on user modifications, adjusts node hierarchical positions, and ensures that the mind map has clear levels; if the user adjustment causes the cause-effect chain to be interrupted, the system uses a path repair algorithm to complete the interrupted path to ensure that the logical chain between nodes is complete.

[0107] The interactive function allows users to freely adjust the map structure and content to meet the personalized needs of different scenarios. The system automatically corrects the node weights and edge relationships adjusted by the user to ensure the logical consistency of the map and the integrity of the causal chain. The visual interface intuitively displays the node and edge relationships, allowing users to quickly understand the logical structure of the map and improve the practicality and readability of the map.

[0108] In the example, a mind map showing the product marketing logic is processed, and the nodes include "sales growth", "user satisfaction improvement" and "market share increase". The user hopes to highlight the importance of "user satisfaction improvement" through interactive adjustment and add annotations to supplement the background information. Processing steps: Visualization interface presentation, node display: "Sales growth" node weight is 0.85, darker color, larger size; "User satisfaction improvement" node weight is 0.65, lighter color, moderate size. Edge relationship display: "User satisfaction improvement → sales growth" edge weight is 0.8, displayed as a thick solid line; "Market share increase → sales growth" edge weight is 0.6, displayed as a thin solid line.

[0109] User adjustment and real-time correction: The user drags the "user satisfaction improvement" node to a position closer to "sales growth", triggering a weight redistribution. After the adjustment, the weight of "user satisfaction improvement → sales growth" increases from 0.8 to 0.85. The user modifies the content of the "user satisfaction improvement" node to "improve user satisfaction (mainly achieved through product optimization)" and adds a note to the node "data comes from the 2024 user feedback survey". The user deletes the edge of "market share increase → sales growth", and the system automatically updates the causal relationship.

[0110] Dynamic update optimization, the system recalculates node weights: The weight of "Increase in user satisfaction" increases from 0.65 to 0.75, and the level is upgraded to the same level as "Increase in sales volume". System repair chain: Due to the deletion of the edge "Increase in market share → Increase in sales volume", the system adds the path "Increase in user satisfaction → Increase in market share".

[0111] The adjusted mind map highlights the core role of "improving user satisfaction" and provides additional background information through annotations. The logical chain is complete and the hierarchy is clear, which facilitates decision-making analysis.

[0112] Preferably, the mind map generation and optimization module verifies the consistency between the mind map and the summary content through a natural language reasoning model, including the following steps: Use the inference model to verify whether the map node content matches the summary semantics, where the inference model uses the node semantic accuracy and summary matching rate as evaluation criteria; Verify whether the edge relationship in the map conforms to the logical chain in the summary content, and mark the nodes or edges that are inconsistent with the verification.

[0113] The mind map generation and optimization module uses a natural language inference model to perform consistency checks on the generated mind maps and summary content, ensuring that the map nodes and edge relationships are consistent with the semantic content and logical chain of the summary. The verification process includes verifying whether the map node content matches the summary semantics, and verifying whether the edge relationships in the map conform to the logical chain in the summary. The model is evaluated by node semantic accuracy and summary matching rate, identifying inconsistent nodes or edges and marking them.

[0114] Use the Natural Language Inference (NLI) model to semantically match map nodes and summary content. The verification process includes the following steps: Generate semantic embedding vectors for the content of map nodes and relevant paragraphs of summaries and , representing its semantic features. The embedding vector is calculated by a pre-trained model (such as XLM-R).

[0115] Calculate the matching rate between node content and summary semantics : , When the matching rate is lower than a threshold (such as 0.8), the node is marked as an “inconsistent node” to indicate that further optimization is required.

[0116] Statistics on the matching rate of all nodes and the semantic accuracy of nodes Defined as: , This value is used to evaluate the semantic consistency of map nodes as a whole.

[0117] Use the reasoning model to verify whether the edge relationship in the map conforms to the logical chain of the summary content: Generate a relationship diagram for the logical chain (such as cause and effect, progressive relationship, etc.) clearly described in the summary as the target chain; compare the edge relationship in the map to verify whether it conforms to the target chain. For example, if the causal relationship described in the summary is "increased user satisfaction → increased sales", it is necessary to verify whether the edge "increased user satisfaction → increased sales" in the map exists, and the edge weight meets the requirements of the strength of the logical chain.

[0118] The logical matching degree of each edge is scored, and the scoring criteria include the logical consistency in the summary, the consistency between the edge weight and the summary relationship strength. Edges with scores below a threshold (such as 0.75) are marked.

[0119] Mark the inconsistent nodes or edges and feed them back to the mind map generation module or summary generation module for adjustment and optimization: Node marking, add annotations to "inconsistent nodes", such as "semantics do not match the summary", and prompt the user to modify or regenerate the node content. Edge marking, add annotations to "inconsistent edges", such as "logical chain interruption", prompting the need to regenerate or adjust the edge weight.

[0120] Through semantic matching and logic chain verification, the map nodes and edge relationships are ensured to be highly consistent with the summary content, so that the map can accurately reflect the core logic of the document. The marking and feedback mechanism supports dynamic adjustment and optimization, and ensures the logical integrity of the map structure through continuous verification and repair. Consistency verification improves the credibility of map content and user trust, providing a more reliable tool for complex document analysis and decision-making.

[0121] Preferably, the mind map generation and optimization module optimizes the mind map structure by generating an adversarial network, comprising the following steps: The generator generates multiple candidate graph structures, each of which contains redistributed node weights and edge relationships; The discriminator evaluates the logical consistency and verifies the semantic coherence of the candidate mind maps, and selects the optimal map structure as the final optimized mind map data.

[0122] The mind map generation and optimization module optimizes the mind map structure through the Generative Adversarial Network (GAN), with the goal of generating an optimal mind map structure with strong logical consistency and semantic coherence. The GAN model consists of a generator and a discriminator: the generator generates multiple candidate mind map structures, each of which contains redistributed node weights and edge relationships; the discriminator evaluates the logical consistency and semantic coherence of these candidate mind maps, and finally selects the optimal mind map as the optimization result. The whole process combines the information of the semantic map and the summary content to ensure that the optimization effect of the mind map meets the logical and semantic requirements.

[0123] The task of the generator is to generate multiple candidate graph structures based on the initial graph and explore the possible allocation of node weights and edge relationships. The specific steps are as follows: The generator uses a random perturbation function to adjust the node weights Make adjustments: , in, is the adjusted node weight; is the original node weight; is the random perturbation value, ; is a hyperparameter for adjusting the weight. The generator adjusts the edge weights Different candidate structures are generated by adding / deleting edges to explore the optimization possibilities of the graph.

[0124] The discriminator verifies the logical consistency and semantic coherence of the candidate maps generated by the generator to ensure that the selected maps meet the system requirements: The discriminator evaluates the consistency of candidate maps based on the logical structure of the map (such as causal chains and hierarchical relationships). The logical consistency scoring formula is: , The "effective chain" refers to a chain that conforms to the logic of cause and effect.

[0125] The discriminator verifies the semantic similarity between the map nodes and the summary content, and calculates the similarity between the node semantic embedding and the summary content embedding: If the semantic similarity and logical consistency scores are both higher than the preset thresholds, the map is determined to be the optimal map.

[0126] The discriminator scores all candidate maps and selects the map with the highest score as the final optimized map based on the comprehensive score (the weighted sum of the logical consistency score and the semantic coherence score).

[0127] Through GAN optimization, the generated mind maps have significantly improved in terms of logical consistency and semantic coherence, and are more in line with user needs and the core logic of the document. The generator explores a variety of mind map structures, and the discriminator selects the optimal solution, achieving a balance between diversity and optimality in mind map optimization. The GAN model can adapt to different document content and summary styles, and continuously improve the map optimization effect through learning.

[0128] Example: A product analysis report generates an initial mind map, which contains three nodes and their relationships: "market demand growth", "user feedback improvement" and "product sales increase". The user wants to optimize the map structure to make it more logical and highlight the key nodes. Processing steps: The generator generates a candidate graph with initial node weights: , , The generator generates candidate weights: Candidate 1: [0.75, 0.72, 0.88]; Candidate 2: [0.82, 0.65, 0.92]. Adjust edge relationships: Candidate 1 deletes the edge "user feedback improvement → market demand growth"; Candidate 2 enhances the edge "market demand growth → product sales increase", and the weight is increased from 0.8 to 0.9.

[0129] Discriminator evaluates candidate map, logical consistency evaluation: logical consistency score of candidate 1 ; Logical consistency score of candidate 2 .

[0130] Semantic similarity score of candidate 1 ; Semantic similarity score of candidate 2 .

[0131] Screening the optimal mind map, the discriminator comprehensive score (weighted sum of logical consistency and semantic coherence): Candidate 1 has a comprehensive score of 0.83; Candidate 2 has a comprehensive score of 0.91. Finally, Candidate 2 is selected as the optimized mind map.

[0132] The optimized mind map highlights the key logic of "market demand growth" and "increase in product sales", deletes redundant edges, and makes the map hierarchy clearer and the semantics more coherent, providing more efficient support for report analysis.

[0133] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. An AI-driven multilingual intelligent document summarization and mind map generation system, characterized in that: include: A multimodal semantic feature generation module is used to receive multimodal document data, extract and fuse semantic features of text content, image data, table data and video subtitle data, and generate unified multimodal semantic feature data; The semantic graph construction module constructs dynamically optimized causal semantic graph data representing the causal relationship between semantic nodes based on multimodal semantic feature data, including causal relationship extraction and multimodal relationship calibration, and generates calibrated causal relationship data; The summary generation module extracts key semantic node content based on the dynamic optimization of causal semantic graph data, uses the dynamic semantic flow modeling method to generate semantic flow paths, and combines the multilingual generation method to generate multilingual summary data; The mind map generation and optimization module generates the initial mind map based on the dynamic optimization of causal semantic graph data and multilingual summary data, and generates the final optimized mind map data through hierarchical relationship optimization and user interaction adjustment.

2. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: In the multimodal semantic feature generation module: The text processing unit performs a word segmentation algorithm on the text content to decompose the semantic units, uses a grammar parsing method to establish a syntactic tree structure, and extracts the text topic through a semantic analysis model; The image processing unit extracts the region boundaries and semantic categories in the image through the image semantic segmentation method, and locates the coordinates and category information of each object in the image in combination with the object detection model; The table processing unit identifies the association between column headers, row headers and cell contents through a table parsing algorithm, and generates table hierarchical structure data; The video subtitle processing unit transcribes the video audio content into text through speech recognition technology, and uses the timing analysis method to mark the timestamp and semantic hierarchy of the subtitle text.

3. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: The semantic graph construction module extracts the causal relationship between semantic nodes by the following method: Use the Bayesian network model to model the statistical correlation of multimodal semantic feature data and establish the potential causal relationship between nodes; The validity of causal relationships is verified using the conditional independence test method, the causal strength between nodes is calculated, and the edges with strengths below a predetermined threshold are removed; The causal relationships that have passed the conditional independence test are recorded as causal relationship data for further construction of the semantic graph.

4. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: The multimodal relationship calibration in the semantic graph construction module includes: A cross-modal relationship graph network is used to optimize the semantic alignment between nodes of different modalities, where the feature representation of the nodes is calibrated by calculating the similarity of semantic vectors between modalities. Based on context consistency, the low-confidence edges in the graph are adjusted, the weights of high-confidence edges are increased, the weights of low-confidence edges are reduced, and invalid edges are removed; The calibrated nodes and edges are constructed as calibrated causal relationship data to dynamically optimize the further generation of causal semantic graphs.

5. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: The summary generation module generates a semantic flow path by a dynamic semantic flow modeling method, including the following steps: Initialize the semantic flow matrix and calculate the semantic flow intensity according to the node weights in the causal semantic graph; The directionality of the semantic flow path is optimized based on the reinforcement learning method, where the target reward function for the semantic flow includes the path semantic coverage, the consistency of the flow direction, and the maximization of the node weight; Extract key node contents along the semantic flow path and generate semantic flow path data for multilingual summary generation.

6. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: The multilingual summary data generated by the summary generation module is verified by a consistency verification method, including the following steps: A multilingual natural language inference model is used to verify the logical consistency of the abstracts in different language versions. The verification model uses the semantic similarity and logical coherence between sentences as evaluation indicators. Mark the inconsistent summary paragraphs, feed the inconsistent parts back to the summary generation module, and regenerate the optimized multilingual summary content.

7. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 1, characterized in that: The mind map generation and optimization module generates the hierarchical structure of the mind map by the following method: Sort the semantic weights of map nodes and divide the priority levels of nodes according to the sorting results; According to the causal relationship between nodes, the lower-level nodes are associated with the higher-level nodes to form a hierarchical map structure; During the map generation process, ensure that the semantic association path between each node and its upper and lower nodes is complete and correct.

8. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 7, characterized in that: The user interaction adjustment function of the mind map generation and optimization module includes: Provides a visual interface for the map, showing the semantic content of nodes and their causal relationship paths; Support users to adjust the position of nodes by dragging them, modify node content or add annotations through input interaction, and correct the adjusted node weights and edge relationships through real-time feedback; Automatically update the map structure modified by the user to the user's adjusted and optimized mind map data.

9. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 8, characterized in that: The mind map generation and optimization module verifies the consistency between the mind map and the summary content through a natural language reasoning model, including the following steps: Use the inference model to verify whether the map node content matches the summary semantics, where the inference model uses the node semantic accuracy and summary matching rate as evaluation criteria; Verify whether the edge relationship in the map conforms to the logical chain in the summary content, and mark the nodes or edges that are inconsistent with the verification.

10. The AI-driven multilingual intelligent document summarization and mind map generation system according to claim 9, characterized in that: The mind map generation and optimization module optimizes the mind map structure by generating adversarial networks, including the following steps: The generator generates multiple candidate graph structures, each of which contains redistributed node weights and edge relationships; The discriminator evaluates the logical consistency and verifies the semantic coherence of the candidate mind maps, and selects the optimal map structure as the final optimized mind map data.

Images