Progressive nested graph visualization method for knowledge-intensive text

Abstract

The invention discloses a knowledge-intensive text-oriented progressive nested graph visualization method, which comprises the following steps of: further decomposing a complex semantic unit only when needed by utilizing an on-demand entity-relationship decomposition module; a nested graph visualization module is used for constructing a decomposition result into nested graph representation; a layout optimization algorithm of structure perception is adopted to enable node distribution to be consistent with a natural reading sequence of the text; a progressive rendering strategy is adopted, incremental presentation is carried out with nodes and sentences as units, and time consistency with a text reading process is kept through smooth and continuous animation transition; in a review link after rendering is finished, context-aware two-way links and entity-driven review interaction are provided, and a user is supported to perform efficient positioning between a nested graph and an original text. According to the method, the complex semantic relation and the context structure can be visually revealed while the text content is completely reserved, the reading cognition burden is reduced, and the understanding efficiency of the knowledge-intensive text is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of text visualization, and particularly relates to a progressive nested graph visualization method for knowledge-intensive text. BACKGROUND

[0002] Text is the most common information carrier and is widely used in daily and professional scenarios. However, reading and understanding text often requires a large cognitive investment, and the reader must follow the author's train of thought and continuously track the relationships between different concepts. For long texts, existing summary and navigation techniques can provide an overall overview or help locate relevant content; but for medium-length texts at the paragraph or chapter level, especially knowledge-intensive texts such as academic papers and scientific treatises, close reading is still needed to accurately grasp their meaning. Such texts contain a large number of professional terms and complex relationships between entities, making it difficult to accurately understand them through browsing or selective reading.

[0003] Knowledge-intensive texts highlight the shortcomings of traditional linear text expression. Linear symbolic sequences lack intuitive structured expression capabilities and are difficult to clearly show the relationships between different entities. At the same time, the lengthy narrative lacks explicit contextual links, forcing the reader to rely on working memory to build an overall semantic model, thereby increasing the burden of understanding. These challenges require new text enhancement techniques that can both preserve the original content and provide assistance during close reading, thereby making up for the lack of summary methods that mainly focus on high-level summaries.

[0004] Existing text enhancement methods usually use graph-based visualization, extracting entities and their relationships from text and displaying them in the form of nodes and edges, providing readers with a structured and clear overview of the content, which has high practical value for general texts. However, when this method is applied to knowledge-intensive texts, it still faces several challenges: first, most existing methods flatten entities and relationships into a single-layer graph structure, either only showing top-level relationships while ignoring the internal composition of clauses, or over-dividing leading to complex and fragmented graph structures; second, existing graph representations are mostly static and overall, making it difficult to align with the natural order of human step-by-step reading. Since readers usually understand content in a top-down, left-to-right order, presenting the complete graph structure at once often causes excessive cognitive burden and affects the semantic continuity between the text and the graph. SUMMARY

[0005] In view of the above, the purpose of the present application is to provide a progressive nested graph visualization method for knowledge-intensive text, which assists in understanding knowledge-intensive text by progressively presenting the entity-relation nested graph of the text.

[0006] To achieve the above-mentioned purpose of the application, the technical solutions provided by the present application are as follows: A progressive nested graph visualization method for knowledge-intensive text, comprising the following steps: By using the on-demand entity-relation decomposition module, sentence-level semantic units are extracted, and complex semantic units are further decomposed into nested entity-relation structures when needed; Based on the decomposition results with nested structures obtained by on-demand entity-relation decomposition, the nested graph visualization module is used to construct the nested graph representation; By using the structure-aware layout optimization algorithm, the force-directed layout of the nested graph is performed according to the position of the entity in the original text as a soft constraint, so as to ensure that the node position is aligned with the reading order of the original text in space; An incremental rendering strategy is adopted to gradually present the nested graph elements in units of sentences and nodes, and smooth and continuous transitions are achieved through animation, so that the expansion process of the nested graph is consistent with the reading order of the text in time; In the review session after rendering, context-aware bidirectional links are provided to enable users to quickly locate and correspond between the nested graph and the original text; In the review session after rendering, entity-driven review interaction is provided to enable users to locate corresponding nodes based on high-frequency entities and highlight related relationships to support review and integration of key information.

[0007] Preferably, the on-demand entity-relation decomposition module comprises: first performing atomic entity-relation decomposition on the input sentence to extract the top-level entity-relation triplets in the sentence, realizing the extraction of sentence-level semantic units; on this basis, further decompose the entities still containing complex semantic structures, and refine the complex semantic units into smaller entities and relationships, to avoid overly rough node representation and improve the clarity of sentence structure expression; when the newly introduced semantic unit overlaps or reuses the existing entity, perform atomic entity-relation decomposition on the new and old entities at the same time, and explicitly establish the cross-sentence context association, so as to maintain the continuity of semantics.

[0008] The atomic entity-relation decomposition is an initial decomposition performed on the input text to extract the topmost entity-relation triplets, and in the decomposition process, isolated entities with no associated relationships and incorrect relationships involving non-existent entities are detected, and the detected errors are corrected to ensure the integrity and accuracy of the decomposition results.

[0009] Preferably, the nested graph visualization module comprises: representing the decomposed entities as nodes, wherein the atomic nodes that are not further decomposed are represented as labeled rounded rectangles, and the composite nodes containing internal subgraphs are represented in the form of containers, and all elements inside the internal subgraphs are enclosed by a closed curve; the decomposed relationships are represented as edges with direction and label, so as to explicitly present the nested decomposition structure and maintain the compactness of visual presentation.

[0010] Preferably, the structure-aware layout optimization algorithm comprises an initial layout strategy and a force-directed layout optimization, wherein the initial layout strategy is used to determine the arrangement of node structures within a sentence to maintain semantic logical relationships, and the force-directed layout optimization iteratively adjusts the positions of the nodes on the basis of the initial layout to enhance structural clarity and maintain alignment with the natural reading order.

[0011] Further preferably, the initial layout strategy adopts a hierarchical layout approach to maintain the logical order of relationships within a single nesting level and arranges the loop nodes as regular polygons when a loop relationship is detected, and the remaining nodes are arranged hierarchically according to their connection relationships with the loop nodes, thereby achieving structural consistency across levels.

[0012] Further preferably, the force-directed layout optimization iteratively calculates the forces acting on the nodes and updates their positions to optimize the overall distribution on the basis of the initial layout, specifically including: in each round of iterative calculation, all nodes are subjected to a comprehensive force and move in space according to the size and direction of the comprehensive force, the internal nodes contained in the composite nodes move as a whole to maintain structural consistency, and the positions of the atomic nodes after moving are discretized to fixed intervals to enhance visual clarity; when the number of iterations exceeds the upper limit or all node positions do not change between the previous and current iterations, the iteration is terminated and the final layout is determined.

[0013] The comprehensive force includes: a connection force acting between connected nodes, similar to a spring generating attraction or repulsion between nodes, maintaining the distance between connected nodes at a preset ideal distance to ensure the spatial rationality of the connection relationship; a containment force acting on the child nodes inside the same composite node, gathering the child nodes towards the center of the composite node, thereby maintaining the compactness inside the composite node and strengthening the semantic association; a repulsion force acting on external nodes mistakenly contained in the composite node region, pushing the node out of the center of the composite node through outward repulsion to prevent misunderstanding due to the inconsistency between the visual presentation and the actual nesting structure in space; an overlap force acting between atomic nodes that are too close to each other, avoiding node overlap through mutual repulsion, thereby enhancing the visual separation and readability of the nodes; and a sentence alignment force acting on nodes belonging to the same sentence, attracting the nodes to the horizontal center line of the sentence to reduce vertical dispersion and maintain clear separation between sentences.

[0014] Preferably, the progressive rendering strategy comprises: introducing sentences and their corresponding nested graphs step by step through user click interaction, reducing the visibility of existing nodes and edges before new content is presented to reduce interference, and organizing nested graphs in columns to maintain the independence of different topics when there is no overlap entity; on this basis, when a new sentence is introduced, the system sequentially performs the following processing steps: when the new sentence shares sub-concepts with the previous text entity, the existing nodes are split, the rectangular boundary of the atomic node gradually expands and evolves into a closed container structure of the composite node during the animation process, and the nodes and edges inside are sequentially presented to reflect the contextual semantic association; then, according to the initial layout strategy, the nodes inside the new sentence are arranged and fixed in position, the force-directed optimization algorithm is applied to the existing nodes to adjust the overall layout, and the unfixed nodes are smoothly moved to the new position; after the position calculation is completed, the incremental nested graph rendering is adopted, and the nested graph of the new sentence is presented in the order from outside to inside and from left to right, the outer structure of the composite node is displayed first, and then the atomic nodes and edges inside are unfolded in sequence to maintain consistency with the text reading order.

[0015] Preferably, the context-aware bidirectional linking dynamically adjusts the display effect when the user interacts with the nested graph or the original text; when the user hovers the cursor over a node in the nested graph, the system reduces the visibility of irrelevant nodes and edges, highlights only the node and its related nodes and edges, and highlights the corresponding text segment in the original text; when the user hovers the cursor over a word in the original text, the system synchronously highlights the related nodes and edges in the nested graph, so that the user can establish a correspondence between the graph and the original text and understand the contextual semantics.

[0016] Preferably, the entity-driven review interaction comprises: according to a preset entity sorting rule, counting and sorting the frequency of occurrence of all entities in the nested graph, and visualizing the top-ranked entities in the sorting result in the form of a column chart; when the user selects a high-frequency entity in the sorting result, the system locates and highlights the entity node and its related nodes and edges in the nested graph, and synchronously highlights the text segment corresponding to the entity in the original text, thereby revealing the contextual semantic association of the entity and supporting efficient information review.

[0017] Preferably, the preset entity sorting rule is based on the connectivity of the entity node in the nested graph, wherein the connectivity value of the composite node is transmitted to the atomic nodes contained therein, so that the sorting result can reflect the importance of the entity in the text semantic structure.

[0018] Compared with the prior art, the present application has at least the following beneficial effects: (1) The nested graph representation based on on-demand entity-relation decomposition can effectively reveal the internal structure of sentences and the cross-sentence context association while preserving the complete content of knowledge-intensive text, avoiding the defects of excessive roughness or fragmentation in existing single-layer graph representation.

[0019] (2) The structure-aware layout optimization algorithm and the progressive rendering strategy are introduced to make the nested graph as consistent as possible with the natural reading order of humans in space and time, reducing cognitive burden and improving overall readability.

[0020] (3) The context-aware bidirectional linking and entity-driven review interaction functions are provided to support efficient positioning and semantic association between text and graph for users, improving the understanding and review efficiency of key information. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 is a flowchart of the progressive nested graph visualization method for knowledge-intensive text in the embodiments of the present application; Figure 2 is a working schematic diagram of the on-demand entity-relation decomposition module and the nested graph visualization module in the embodiments of the present application; Figure 3 is a schematic diagram of the principle of the comprehensive force involved in the structure-aware layout optimization algorithm in the embodiments of the present application; Figure 4 is a flowchart and effect schematic diagram of the progressive rendering strategy in the experimental example of the present application; Figure 5 is a schematic diagram of the context-aware bidirectional linking interaction effect in the embodiments of the present application; Figure 6 is a schematic diagram of the entity-driven review interaction effect in the embodiments of the present application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the protection scope of the present application.

[0024] The inventive concept of this invention is to address the problems of existing graph-based text visualization methods when processing knowledge-intensive text, such as the difficulty of single-layer graphs simultaneously representing high-level relationships and detailed structures, and the inconsistency between static representations and human reading processes. Therefore, this invention proposes a progressive nested graph visualization method for knowledge-intensive text, assisting users in more efficiently grasping complex semantic relationships and contextual structures while fully preserving the text content, thereby improving the understanding of knowledge-intensive text.

[0025] Figure 1 This is a flowchart illustrating a progressive nested graph visualization method for knowledge-intensive text in an embodiment of the present invention. Figure 1 As shown, the embodiment provides a progressive nested graph visualization method for knowledge-intensive text, including the visualization of the reading process ( Figure 1 (A) and review process ( Figure 1 Support for (B) includes the following steps: A1 utilizes the on-demand entity-relation decomposition module to extract sentence-level semantic units and further decompose complex semantic units into nested entity-relation structures when needed.

[0026] Specifically, such as Figure 2 As shown in Figure B, the on-demand entity-relation decomposition module first performs atomic entity-relation decomposition on the input sentence, extracting the top-level entity-relation triples from the sentence. Figure 2 China B1 and Figure 2 (A); Based on this, when an entity is still structurally a sentence, it indicates that it contains a complex semantic structure. This module will further decompose it into smaller entities and relations (A). Figure 2 (B2), to avoid overly coarse node representations and improve the clarity of sentence structure expression.

[0027] Furthermore, this module will detect the decomposed entities to determine whether they have a noun substring with the same concept as the entities mentioned earlier. Specifically, when the sub-concept of an entity in this sentence has the same semantic consistency with a certain entity mentioned above, this module will further decompose the entity in this sentence. Figure 2 (B3) enables the entity to correspond with existing entities in the graph structure; when an entity in this sentence has the same semantic meaning as a sub-concept of an entity mentioned earlier, this module will further decompose the entity mentioned earlier ( Figure 2 (B4), enabling it to correspond to newly introduced entities; in some cases, the above two types of situations may occur simultaneously, for example, when the newly appearing entity in this sentence partially overlaps semantically with the entity mentioned earlier, the module will decompose both types of entities simultaneously ( Figure 2B3 and B4) to ensure the continuity of the context.

[0028] Specifically, the atomic entity-relation decomposition module performs initial decomposition on the input text to extract top-level entity-relation triples in the sentence. In a specific implementation, a large language model GPT-4o is used to label the input text to generate an initial result containing only top-level entity-relation structures without further decomposition. To improve the decomposition quality, the module also detects two common errors, including isolated entities not connected to any relationship and incorrect relationships involving non-existent entities. When such inconsistencies are detected, the previously returned results and detected errors are used as context to call GPT-4o for self-correction, thereby ensuring the completeness and accuracy of the decomposition results.

[0029] A2, based on the decomposition results with nested structures obtained by on-demand entity-relation decomposition, a nested graph visualization module is used to construct a nested graph representation.

[0030] Specifically, as shown in Figure 2 C, the nested graph visualization module visualizes the decomposition results with nested structures. All entities and relationships in each sentence are uniformly wrapped by a gray dashed line box, and the sentence number is attached for identification to distinguish the semantic scope of different sentences (C1). Figure 2 The entity nodes obtained by decomposition are divided into two categories: atomic entities that are not further decomposed are represented by a rounded rectangle with a text label (C2), which visually presents the basic semantic units in the sentence; composite entities containing internal subgraphs are represented in the form of a container, which encloses the internal entities and relationships with a closed curve and forms a boundary in the visual to reflect the nested structure (C3). Figure 2 Figure 2 The relationships obtained by decomposition are represented by edges with direction and label, which clearly show the semantic connection between entities (C4). This visualization method not only explicitly shows the nested structure in semantics, but also keeps the overall layout compact to avoid excessive fragmentation of the graph. Figure 3

[0031] A3, using a structure-aware layout optimization algorithm, the position of the entity in the original text as a soft constraint, force-directed layout of the nested graph to ensure that the node position is aligned with the reading order in space.

[0032] Specifically, the structure-aware layout optimization algorithm includes an initial layout strategy and a force-directed layout optimization, where the initial layout strategy is used to determine the node structure arrangement within a sentence to maintain semantic logical relationships, and the force-directed layout optimization iteratively adjusts the node position based on the initial layout to enhance the structure clarity and maintain alignment with the natural reading order. ​​

[0033] Further specifically, the initial layout strategy maintains the logical order of relationships in a hierarchical layout within a single nesting level, and arranges the loop nodes as regular polygons when a loop relationship is detected, and the remaining nodes are hierarchically laid out according to the connection relationship with the loop nodes, thereby achieving structural consistency across levels.

[0034] Further specifically, the force-directed layout optimization iteratively calculates the forces acting on the nodes and updates their positions to optimize the overall distribution based on the initial layout. In each iteration, all nodes are subjected to a comprehensive force and move in space according to the size and direction of the comprehensive force. When the composite nodes move, the internal nodes contained therein are displaced as a whole to maintain structural consistency. The positions of the atomic nodes after moving are then discretized to fixed intervals to enhance visual clarity. When the number of iterations exceeds the upper limit or all node positions remain unchanged between two consecutive iterations, the iteration is terminated and the final layout is determined.

[0035] As shown in Figure 3 , the comprehensive force includes the following five types: (1) Connection force (A in Figure 3 ), which acts between connected nodes, similar to a spring that generates attraction or repulsion between nodes, maintaining the distance between connected nodes at a predetermined ideal distance to ensure spatial rationality of the connection relationship; (2) Inclusion force (B in Figure 3 ), which acts on the child nodes within the same composite node, gathering the child nodes towards the center of the composite node, thereby maintaining the compactness within the composite node and strengthening the semantic association; (3) Repulsion force (C in Figure 3 ), which acts on external nodes mistakenly included within the composite node region, pushing the node out of the center of the composite node through outward repulsion, preventing misinterpretation due to inconsistency between visual presentation and actual nesting structure in space; (4) Overlapping force (D in Figure 3 ), which acts between atomic nodes that are too close to each other, repelling each other to avoid node overlap, thereby enhancing the visual separation and readability of the nodes; (5) Sentence alignment force (E in Figure 4 ), which acts on nodes belonging to the same sentence, attracting the nodes to the horizontal center line of the sentence to reduce vertical dispersion and maintain clear separation between sentences.

[0036] A4, adopts a progressive rendering strategy, presenting nested graph elements in units of sentences and combining nodes step by step, and achieves smooth and continuous transition through animation, making the unfolding process of the nested graph consistent with the reading order of the text in terms of time.

[0037] Specifically, the progressive rendering strategy first employs a click-driven, sentence-level presentation method. Users gradually navigate to the next sentence and its corresponding nested graph through clicks. Before new content is presented, the system reduces the visibility of existing nodes and edges to minimize distractions and highlight newly introduced content. In particular, for sentences that do not involve overlapping entities, their corresponding nested graphs are organized by columns to maintain the independence of different themes, reduce visual clutter as text grows, and help users identify thematic changes and track shifts in information focus during the progressive browsing process.

[0038] Based on this, when a new sentence is introduced, the system will perform the following processing steps in sequence: First, when the new sentence shares the same sub-concepts as the preceding entity, such as... Figure 4 As shown in Figure A, the system splits existing related nodes on the canvas to establish new semantic connections. During this process, the rectangular boundaries of the atomic nodes gradually expand and evolve into a closed container structure of composite nodes through animation. Figure 4 (A1), whose internal nodes and edges are then presented in sequence ( Figure 4 (A2), thus intuitively reflecting the semantic correspondence of context in the graph structure. Subsequently, the system arranges and fixes the positions of nodes within the new sentence according to the initial layout strategy, and applies a force-oriented optimization algorithm to the existing nodes on the canvas to adjust the layout at the overall level. Unfixed nodes on the canvas are smoothly moved to their new positions, such as... Figure 4 As shown in Figure B, this ensures the overall structure remains harmonious and clear. After completing the above steps, as shown... Figure 4 As shown in Figure C, the system uses incremental nested graph rendering, introducing nested graphs of new sentences in a rendering order from the outside to the inside and from left to right. Figure 5 In the example of C1 and C2, the external structure of the composite node is presented first, and then its internal atomic nodes and edges are gradually expanded to maintain consistency with the text reading order.

[0039] B1 provides context-aware bidirectional links in the post-rendering review phase, enabling users to quickly locate and correspond between nested diagrams and the original text.

[0040] Specifically, such as Figure 5 As shown, when a user hovers the cursor over a node in a nested graph, regardless of whether that node is an atomic node ( Figure 5 (A) or composite node ( Figure 5 In the case of node B, the system will reduce the visibility of irrelevant nodes and edges, and only highlight the node in question. Figure 5 (A1 and B1) and their directly related nodes and edges ( Figure 5 (See A2 and B2). Simultaneously, the corresponding text segments in the original text will be highlighted. Figure 6A3 and B3), so as to enable the user to quickly locate the relevant paragraph and confirm the context details. Conversely, when the user hovers the cursor over a certain word or phrase in the original text, the system highlights the corresponding node and its related connections in the nested graph, thereby providing a structured relationship view. In this way, when the user is reading the original text, the user can immediately see the semantic relationship network in which the word is located, thereby assisting in understanding the role and function of the word in the context.

[0041] B2, in the review session after the rendering is completed, entity-driven review interaction is provided, so that the user can locate the corresponding node based on the high-frequency entity and highlight the relevant relationship, so as to support review and integration of key information.

[0042] Specifically, as shown in Figure 6 , the system counts and sorts the frequency of occurrence of all entities in the nested graph according to a preset entity sorting rule, and visualizes the top-ranked entities in the sorting result in the form of a column chart; when the user selects a high-frequency entity in the sorting result, the system locates and highlights the entity node and its related nodes and edges in the nested graph Figure 6 , and simultaneously highlights the text segment corresponding to the entity in the original text ​ , thereby revealing the context semantic association of the entity and supporting efficient information review.

[0043] Specifically, the preset entity sorting rule is to sort the entities based on the connection degree of the entity nodes in the nested graph, wherein the connection degree value of the composite node is transmitted to the atomic nodes contained in the composite node, so that the sorting result can reflect the importance of the entity in the text semantic structure.

[0044] In summary, the present application provides a progressive nested graph visualization method for knowledge-intensive text, which realizes the gradual decomposition from the top-level triple to the fine-grained nested structure by introducing an on-demand entity-relation decomposition module, and intuitively presents the semantic hierarchy in combination with the nested graph visualization module; a structure-aware layout optimization algorithm is used to make the graph structure consistent with the reading order of the original text in space; meanwhile, a progressive rendering strategy is used to synchronize with the unfolding process of the text in the time dimension, thereby reducing the cognitive burden of the user; in the review session, context-aware bidirectional linking and entity-driven review interaction are also provided, so as to support efficient positioning and integration of the user between the nested graph and the original text. Through the above technical solutions, the present application can help the user to more clearly understand the complex semantic relationship and context structure while retaining the content of the text, thereby improving the understanding efficiency of the knowledge-intensive text, and having a wide application prospect.

[0045] The above detailed description of the specific embodiments of the present application has described the technical solutions and beneficial effects of the present application, and it should be understood that the above description is only the most preferred embodiment of the present application and is not intended to limit the present application. Any modifications, supplements and equivalent replacements made within the principle range of the present application shall be included in the protection range of the present application.

Claims

1. A progressive nested graph visualization method for knowledge-intensive text, characterized in that, Includes the following steps: By utilizing the on-demand entity-relation decomposition module, sentence-level semantic units are extracted, and complex semantic units are further decomposed into nested entity-relation structures when needed. Based on the nested structure of the decomposition results obtained from the on-demand entity-relationship decomposition, a nested graph representation is constructed using the nested graph visualization module. Using a structure-aware layout optimization algorithm, the nested graph is force-guided to ensure that the node positions are spatially aligned with the reading order of the original text, based on the position of the entity in the original text as a soft constraint. A progressive rendering strategy is adopted, which presents nested graph elements step by step, sentence by sentence, and uses animation to achieve a smooth and continuous transition, so that the unfolding process of the nested graph is consistent with the reading order of the text. In the post-rendering review phase, context-aware bidirectional links are provided, enabling users to quickly locate and correspond between nested diagrams and the original text; The post-rendering review phase provides entity-driven review interaction, enabling users to locate corresponding nodes based on high-frequency entities and highlight related relationships.

2. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The on-demand entity-relationship decomposition module includes: First, atomic entity-relation decomposition is performed on the input sentence to extract the top-level entity-relation triples, thus achieving the extraction of sentence-level semantic units. On this basis, entities that still contain complex semantic structures are further decomposed, refining the complex semantic units into smaller entities and relations. When newly introduced semantic units overlap or reuse entities that have appeared in the previous text, atomic entity-relation decomposition is performed on both the new and old entities to explicitly establish cross-sentence contextual associations, thereby maintaining semantic continuity. In the process of atomic entity-relation decomposition, isolated entities with no correlation and erroneous relations involving entities that do not exist are detected, and the detected errors are self-corrected.

3. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The nested graph visualization module includes: When constructing a nested graph representation, the entities obtained from decomposition are represented as nodes, where atomic nodes that are not further decomposed are represented as labeled rounded rectangles, and composite nodes containing internal subgraphs are represented as containers, with all elements of their internal subgraphs enclosed by closure curves; the relationships obtained from decomposition are represented as edges with directions and labels, thus explicitly presenting the nested decomposition structure while maintaining visual compactness.

4. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The structure-aware layout optimization algorithm includes an initial layout strategy and a force-guided layout optimization. The initial layout strategy is used to determine the arrangement of nodes within a sentence to maintain semantic logical relationships. The force-guided layout optimization iteratively adjusts the node positions based on the initial layout to enhance structural clarity and maintain alignment with the natural reading order.

5. The progressive nested graph visualization method for knowledge-intensive text according to claim 4, characterized in that, The initial layout strategy maintains the logical order of relationships by using a hierarchical layout within a single nested level. When a cyclic relationship is detected, the cyclic nodes are arranged into regular polygons, and the remaining nodes are laid out hierarchically according to their connection with the cyclic nodes, thereby achieving structural consistency across levels.

6. The progressive nested graph visualization method for knowledge-intensive text according to claim 4, characterized in that, The force-guided layout optimization optimizes the overall distribution by iteratively calculating the forces acting on nodes and updating their positions based on the initial layout. Specifically, in each round of iterative calculation, all nodes are subjected to a combined force and move in space according to the magnitude and direction of the combined force. When a composite node moves, its internal nodes are displaced as a whole to maintain structural consistency. The positions of atomic nodes after moving are discretized to fixed intervals to enhance visual clarity. When the number of iterations exceeds the upper limit or the positions of all nodes in two consecutive iterations no longer change, the iteration is terminated and the final layout is determined. The combined forces include: Connectivity force acts between connected nodes to maintain the distance between them at a preset ideal distance, thus ensuring the spatial rationality of the connection relationship. Inclusion force acts on child nodes within the same composite node, bringing the child nodes together toward the center of the composite node, thereby maintaining the compactness within the composite node and strengthening semantic associations. Repulsive force acts on external nodes that are mistakenly included in the composite node area, pushing the node out of the center of the composite node through outward repulsion, thus preventing misunderstandings caused by inconsistencies between the spatial visual presentation and the actual nested structure. Overlapping force acts between atomic nodes that are too close to each other, preventing node overlap by repelling each other, thereby enhancing the visual separation and readability of nodes; Sentence alignment force acts on nodes belonging to the same sentence, pulling the nodes to the horizontal center line of the sentence to reduce vertical dispersion and maintain clear separation between sentences.

7. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The progressive rendering strategy includes: gradually introducing sentences and their corresponding nested graphs through user clicks; reducing the visibility of existing nodes and edges to minimize interference before new content is presented; and organizing the nested graphs by columns to maintain the independence of different themes when no overlapping entities are involved. Based on this, when a new sentence is introduced, the system sequentially executes the following processing steps: when the new sentence shares sub-concepts with preceding entities, existing nodes are split; the rectangular boundaries of atomic nodes gradually expand and evolve into closed container structures of composite nodes during the animation process, with their internal nodes and edges presented sequentially to reflect contextual semantic relationships; subsequently, the nodes inside the new sentence are arranged and their positions fixed according to the initial layout strategy; a force-oriented optimization algorithm is applied to existing nodes to adjust the overall layout, and unfixed nodes are smoothly moved to new positions; after position calculation, incremental nested graph rendering is used, presenting the nested graph of the new sentence in an order from outside to inside and from left to right, first displaying the external structure of composite nodes, and then sequentially expanding the atomic nodes and edges to maintain consistency with the text reading order.

8. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The context-aware bidirectional links dynamically adjust the display effect when the user interacts with the nested graph or the original text. When the user hovers the cursor over a node in the nested graph, the system reduces the visibility of irrelevant nodes and edges, highlighting only the node and its related nodes and edges, while simultaneously highlighting the corresponding text segment in the original text. When the user hovers the cursor over a word in the original text, the system simultaneously highlights the related nodes and edges in the nested graph, enabling the user to establish a correspondence between the graph and the original text and understand the contextual semantics.

9. The progressive nested graph visualization method for knowledge-intensive text according to claim 1, characterized in that, The entity-driven review interaction includes: statistically analyzing and sorting the frequency of occurrence of all entities in the nested graph according to preset entity sorting rules, and visually displaying the top-ranked entities in the sorting results in the form of a bar chart; when the user selects a high-frequency entity in the sorting results, the system locates and highlights the entity node and its related nodes and edges in the nested graph, and simultaneously highlights the text fragment corresponding to the entity in the original text, thereby revealing the contextual semantic relationship of the entity and supporting efficient information review.

10. The progressive nested graph visualization method for knowledge-intensive text according to claim 9, characterized in that, The preset entity sorting rule sorts entities based on the connectivity of entity nodes in the nested graph, wherein the connectivity value of composite nodes is passed to the atomic nodes they contain, so that the sorting result can reflect the importance of entities in the semantic structure of the text.

Images