A nebula chart hierarchical progressive visualization method for large-scale spatiotemporal knowledge graph

Abstract

The present application belongs to the technical field of nebula chart hierarchical progressive visualization method, and particularly relates to a nebula chart hierarchical progressive visualization method for large-scale spatiotemporal knowledge graph, which adopts knowledge graph input data preprocessing, adaptive hierarchical clustering based on spatiotemporal information, and front-end visualization and interactive control.

Description

Technical Field

[0001] This invention belongs to the technical field of hierarchical progressive visualization methods for nebula graphs, and particularly relates to a hierarchical progressive visualization method for nebula graphs for large-scale spatiotemporal knowledge graphs. Background Technology

[0002] Existing visualization methods are prone to severe node overlap and edge intersection when dealing with large-scale nodes, resulting in visual congestion. For example, when the number of nodes in a graph exceeds 100,000, traditional scatter plot visualizations cause nodes to be densely stacked in space, making it impossible to distinguish the location information and relationships of individual nodes. On the other hand, simplifying the display of edges will lose key topological structures, making it impossible for users to accurately identify the dependency logic between entities, which greatly reduces the readability and analytical value of the graph.

[0003] The core value of spatiotemporal knowledge graphs lies in the synergistic expression of spatiotemporal dimensions and topological relationships. However, existing methods often suffer from dimensional fragmentation: some methods focus only on visualizing spatial distribution, ignoring the dynamic evolution of the temporal dimension; others struggle to simultaneously present the topological relationships between entities, resulting in a failure to collaboratively display spatiotemporal patterns and relational logic. For example, in traffic scheduling scenarios, existing methods cannot simultaneously and clearly present the congestion status of road nodes, the spatial distribution of road nodes, and the connectivity between roads within different time periods, affecting the accuracy of scheduling decisions.

[0004] To alleviate visual congestion, some methods attempt to simplify the knowledge graph through hierarchical display. However, these methods often employ fixed hierarchical division rules, which cannot adapt to the uneven node density in spatiotemporal knowledge graphs. For example, in visualizing traffic nodes in urban core areas and suburbs, the core area has a high node density, and a fixed hierarchical division can lead to over-aggregation of nodes within clusters in the core area, resulting in the loss of detailed information; conversely, the suburbs have a low node density, leading to insufficient clusters and broken relationships. Furthermore, existing hierarchical methods cannot dynamically adjust the granularity of the hierarchy based on the spatiotemporal relationships of nodes, resulting in a disconnect between the hierarchical division and the actual data patterns.

[0005] Effective data preprocessing and clustering are fundamental to large-scale geographic graph visualization, but existing methods suffer from two shortcomings: First, the node embedding process fails to fully integrate topological structure and spatiotemporal information. Some methods embed nodes solely based on topological structure, ignoring the impact of spatiotemporal dimensions on node correlation; others, while considering spatiotemporal information, fail to balance topological and spatiotemporal features through scientific weight allocation, resulting in embedding vectors that cannot accurately reflect the comprehensive correlation of nodes. Second, clustering algorithms lack adaptability. Traditional algorithms such as DBSCAN and K-Means require manual setting of key parameters and cannot dynamically adjust clustering strategies based on the spatiotemporal volume and density of clusters, leading to poor stability of clustering results and failing to provide reliable support for subsequent hierarchical visualization.

[0006] Users often need to use interactive operations such as panning and zooming to obtain information at different granularities when exploring spatiotemporal knowledge graphs. However, the interactive response mechanism of existing visualization methods has defects: on the one hand, when users pan or zoom, the backend data query and frontend rendering are not synchronized, which can easily lead to data loading delays or information display gaps; on the other hand, it is impossible to dynamically adjust the displayed hierarchical details according to the zoom level. For example, when zooming at a small scale, the top-level abstract cluster is not displayed, making it impossible for users to grasp the global rules; when zooming at a large scale, the details of the underlying nodes are not loaded, making it impossible for users to deeply analyze local relationships, which seriously affects the user's exploration efficiency. Summary of the Invention

[0007] The purpose of this invention is to address the aforementioned technical problems by providing a hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs.

[0008] In view of this, the present invention provides a hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs, comprising the following steps:

[0009] Step 1: Preprocess the input data using a knowledge graph;

[0010] Step 2: Adaptive hierarchical clustering based on spatiotemporal information;

[0011] Step 3: Perform front-end visualization and interactive control.

[0012] Preferably, the specific input data preprocessing in step one includes the following steps:

[0013] Step S11: Define the knowledge graph Where V is the set of nodes, and each node Includes spatiotemporal information: location coordinates and timestamp ;

[0014] Step S12: Extract each node using a graph embedding method. Topological embedding vector The embedded dimension d is selected from a preset set of dimensions through grid search to optimize the accuracy of link prediction;

[0015] Step S13: Standardize the node's position coordinates and timestamp using the z-score normalization method to obtain the spatiotemporal embedding vector:

[0016] ;

[0017] in, These are the standard deviations of the location coordinates x and y, and the timestamp t, respectively.

[0018] Step S14: Calculate the comprehensive embedding for each node:

[0019] ;

[0020] in, It is a hyperparameter, and its value range is determined through Sobol sensitivity analysis. It is then optimized by cross-validation grid search, and the objective function is the Silhouette score.

[0021] Preferably, step two of the adaptive hierarchical clustering further includes the following steps:

[0022] Step S21: Define the composite distance between nodes:

[0023] ;

[0024] in It's weight. ;

[0025] The spacetime Euclidean distance, The time scaling factor is calculated based on the spatiotemporal scale matching theory, and the formula is as follows: ;

[0026] Step S22: Perform hierarchical clustering of the knowledge graph based on density: First calculate the density of each node. Local density ,in The initial threshold is determined by analyzing the inflection point using a k-distance graph.

[0027] Then adaptively adjust the neighborhood radius K initial clusters were obtained. ;

[0028] Step S23: For each initial cluster Perform top-down hierarchical partitioning: calculate cluster centroids. Starting from the second layer, representative nodes are selected based on the centroid, and the clusters are divided into... Subclusters;

[0029] If cluster density The spatiotemporal volume of the cluster is approximated by the convex hull volume. If the density threshold is reached, the number of levels is increased; otherwise, the partitioning process stops, ultimately resulting in clusters. hierarchical set And satisfy ;

[0030] Step S24: Minimize the intra-class distance and hierarchical consistency loss, the loss function is... ,in For the first The hierarchical center of the layer.

[0031] Preferably, step three, the front-end visualization and interactive control, specifically includes the following steps:

[0032] Step S31: Listen for user panning operations and obtain the current visible area coordinate range:

[0033] Send a query to the backend database to obtain the required data. node subset ;

[0034] Step S32: Listen for user zoom operations and obtain the current zoom level. Combined with preset threshold Control the display hierarchy: If Show top layer ;

[0035] like Display nodes from level l to level l+1; if z This displays all levels.

[0036] Preferably, the preset dimension set mentioned in step S12 is {64, 128, 256}. When d=128, the link prediction accuracy on the river basin knowledge graph dataset is optimal, with an MRR value of 0.35, which is higher than 0.28 when d=64 and 0.32 when d=256.

[0037] The steps described in step S14 The value range is [0.1, 0.9]. In the experiment of spatiotemporal knowledge graph based on river basin, when The Silhouette score is optimal at 0.7, which is 0.68.

[0038] Preferably, the value range of β in step S21 is [0.1, 0.9]. In the knowledge graph experiment based on administrative divisions, the ARI value is optimal when β = 0.6, which is 0.75; and in this experiment, the clustering stability is highest when γ = 0.25.

[0039] The range of θ in step S23 is [0.1, 1], with the optimal value being 0.5.

[0040] Preferably, the hierarchical threshold mentioned in step S32 Through user interaction testing, it was determined that the information density of the visual interface is moderate at different zoom levels, without being overly crowded or overly sparse.

[0041] The graph embedding method described in step S12 is an unsupervised embedding algorithm based on graph structure, including but not limited to Node2Vec, DeepWalk, or GCN, to effectively capture the topological structure information of the knowledge graph.

[0042] Preferably, the increment δ of the number of sub-clusters in step S23 is a positive integer, and the specific value is adaptively adjusted according to the cluster size to ensure that each sub-cluster contains a sufficient number of nodes to maintain clustering stability.

[0043] Preferably, the backend database in step S31 uses a database system that supports spatiotemporal indexes, including but not limited to PostgreSQL with PostGIS extensions, to improve the node subset. Query efficiency.

[0044] Preferably, the initial number of clusters K in S22 is determined by combining the elbow rule with clustering effect evaluation to ensure that the clustering results avoid both fragmentation caused by too many clusters and excessive information aggregation caused by too few clusters.

[0045] The beneficial effects of this invention are:

[0046] By preprocessing and persisting layout calculations, and implementing an on-demand loading strategy based on coordinate range and scaling on the front end, the amount of computation and data transfer required for each rendering is significantly reduced. This effectively solves performance issues such as stuttering and slow loading of large-scale maps in browsers, achieving a smooth, real-time translation and scaling interactive experience.

[0047] It provides a progressive exploration path from macro to micro. Users can first understand the overall structure of the knowledge graph through high-level abstract nodes, and then gradually delve into communities or entities of interest by zooming in to view their detailed information. This hierarchical view reduces the cognitive burden caused by information overload, enabling users to discover hidden connections and patterns in the knowledge graph more efficiently, thus improving insight efficiency.

[0048] By tightly integrating backend data abstraction and data storage with frontend layout calculations and interactive rendering, a highly efficient, scalable, and logically coherent overall solution is formed. This makes it more robust and scalable than existing solutions, enabling it to handle future challenges related to larger-scale knowledge graphs. Attached Figure Description

[0049] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0051] It should be noted that all directional and positional terms used in this invention, such as "up," "down," "left," "right," "front," "back," "vertical," "horizontal," "inner," "outer," "top," "lower," "lateral," "longitudinal," and "center," are only used to explain the relative positional relationships and connections between components in a specific state (as shown in the accompanying drawings). They are merely for the convenience of describing the invention and do not require the invention to be constructed and operated in a specific orientation; therefore, they should not be construed as limitations on the invention. Furthermore, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0052] In the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0053] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0054] This invention provides a hierarchical, progressive visualization method for large-scale spatiotemporal knowledge graphs. Its overall architecture includes a backend data processing and storage module and a frontend visualization and interaction module, achieving smooth display of massive knowledge graphs through close system integration. Each step will be described in detail below.

[0055] Step 1: Preprocessing of knowledge graph input data;

[0056] knowledge graph ,in It is a set of nodes, each node Includes spatiotemporal information: location coordinates and timestamp .

[0057] Extract each node using graph embedding methods. Embedded vector It captures the structural information of the graph. The embedded dimensions... Selecting from {64,128,256} using a grid search, experiments show that when... At that time, the best link prediction accuracy was achieved on the river basin knowledge graph dataset (MRR=0.35, higher than the previous result). 0.28 and (0.32).

[0058] After standardizing the location and time, they are concatenated as follows: ,in This is the standard deviation of the corresponding dimension. The standard deviation is calculated based on the z-score standardization method, ensuring that the mean of each dimension is 0 and the variance is 1, thus avoiding the dominance of scale unevenness in distance calculation.

[0059] Calculate the comprehensive embedding for each node ,in It is a hyperparameter that controls the weights of topology and spatiotemporal information. The determination was based on Sobol sensitivity analysis, with values ​​ranging from [0.1, 0.9], and optimized using a 5-fold cross-validation grid search. The objective function was the Silhouette score. In the experiment of a spatiotemporal knowledge graph based on a river basin, The best score was 0.68.

[0060] Step 2: Adaptive hierarchical clustering based on spatiotemporal information

[0061] Define the overall distance between nodes:

[0062] ;

[0063] in It's weight. The spacetime Euclidean distance, It is the time scaling factor. The value range is [0.1, 0.9], determined through grid search and ARI evaluation in a knowledge graph experiment based on administrative divisions. =0.6 is optimal (ARI=0.75). The theoretical basis is that the convex combination of weighted Euclidean distances ensures that the distance metric satisfies the triangle inequality. The determination is based on the spatiotemporal scale matching theory, and the calculation formula is:

[0064] ;

[0065] Experiments using knowledge graph data based on administrative divisions show that when... Time-based clustering offers the highest stability, preventing the time dimension from becoming dominant or ignored.

[0066] Hierarchical clustering of knowledge graphs based on density, with neighborhood radius parameter The minimum number of points, minPts, is adaptively adjusted based on the local density. For each node... Estimate local density:

[0067] ;

[0068] in That's the initial threshold. Then, it's adaptively adjusted. This addresses the issue of uneven density, ultimately resulting in K initial clusters of different sizes. ,in The method for setting up the graph is k-distance graph analysis, sorting the distances of the k-th nearest neighbors and taking... As an inflection point. Experiments show that when The cluster number is most stable at the inflection point, and its impact on the final clustering result is: too small a value... This leads to cluster fragmentation, and excessively large clusters lead to merging.

[0069] For each cluster A top-down hierarchical partitioning is performed, with a total of L layers (L being the preset maximum number of layers). Starting from the second layer, representative nodes are selected from the central node of the upper layer, and the centroid of the cluster is calculated:

[0070] ;

[0071] The lower-level nodes are further expanded through pseudo-label clustering: The nodes in the middle are divided into sub-clusters The center of each sub-cluster serves as the representative node of that layer.

[0072] If cluster density If it does, then increase the level; otherwise, stop. Density ,in It is the spatiotemporal volume of the cluster (approximate to the convex hull volume). The stability index is determined by cross-validation optimization, with a value range of [0.1, 1], and is selected by minimizing the inter-level variance. =0.5.

[0073] Minimize intra-class distance and hierarchy consistency loss:

[0074] ;

[0075] in It is the first The hierarchical center of the layer.

[0076] Each cluster Divided into hierarchical sets ,in (The underlying layer contains all nodes).

[0077] The front-end interface monitors the user's panning (Pan) operation in real time to obtain the coordinate range of the currently visible area. Use this range as the query condition to send a request to the backend database: query results that meet the condition. nodes Returns a subset of nodes. .

[0078] Simultaneously monitor zoom operations and the current zoom level. According to the preset threshold The display hierarchy is determined if... Show top layer (Abstract View); If This displays the nodes from level l to level l+1; All levels are displayed at that time.

[0079] This solution was tested on a knowledge graph of Chinese administrative divisions. The results showed that this solution outperformed existing knowledge graph visualization solutions Citespace and Neo4j in all aspects.

[0080]

[0081] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs, characterized in that: Includes the following steps: Step 1: Preprocess the input data using a knowledge graph; Step 2: Adaptive hierarchical clustering based on spatiotemporal information; Step 3: Perform front-end visualization and interactive control; Step one involves the following steps in the specific input data preprocessing: Step S11: Define the knowledge graph Where V is the set of nodes, and each node Includes spatiotemporal information: location coordinates and timestamp ; Step S12: Extract each node using a graph embedding method. Topological embedding vector The embedded dimension d is selected from a preset set of dimensions through grid search to optimize the accuracy of link prediction; Step S13: Standardize the node's position coordinates and timestamp using the z-score normalization method to obtain the spatiotemporal embedding vector: ； in, These are the standard deviations of the location coordinates x and y, and the timestamp t, respectively. Step S14: Calculate the comprehensive embedding for each node: ； in, These are hyperparameters. Their value range is determined through Sobol sensitivity analysis, and then optimized using cross-validation grid search. The objective function for optimization is the Silhouette score. Step two of adaptive hierarchical clustering specifically includes the following steps: Step S21: Define the composite distance between nodes: ； in It's weight. ; The spacetime Euclidean distance, The time scaling factor is calculated based on the spatiotemporal scale matching theory, and the formula is as follows: ; Step S22: Perform hierarchical clustering of the knowledge graph based on density: First calculate the density of each node. Local density ,in The initial threshold is determined by analyzing the inflection point using a k-distance graph. Then adaptively adjust the neighborhood radius K initial clusters were obtained. ; Step S23: For each initial cluster Perform top-down hierarchical partitioning: calculate cluster centroids. Starting from the second layer, representative nodes are selected based on the centroid to divide the clusters into... Subclusters; If cluster density The spatiotemporal volume of the cluster is approximated by the convex hull volume. If the density threshold is reached, the number of levels is increased; otherwise, the partitioning process stops, ultimately resulting in clusters. hierarchical set And satisfy ; Step S24: Minimize the intra-class distance and hierarchical consistency loss, the loss function is... ,in For the first The layer's hierarchy center; Step three, front-end visualization and interactive control, specifically includes the following steps: Step S31: Listen for user panning operations and obtain the current visible area coordinate range: Send a query to the backend database to obtain the required data. node subset ; Step S32: Listen for user zoom operations and obtain the current zoom level. Combined with preset threshold Control the display hierarchy: If Show top layer ; like Display nodes from level l to level l+1; if This displays all levels.

2. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 1, characterized in that: The preset dimension set mentioned in step S12 is {64, 128, 256}. When d=128, the link prediction accuracy on the river basin knowledge graph dataset is optimal, with an MRR value of 0.35, which is higher than 0.28 when d=64 and 0.32 when d=256. The steps described in step S14 The value range is [0.1, 0.9]. In the experiment of spatiotemporal knowledge graph based on river basin, when The Silhouette score is optimal at 0.7, which is 0.

68.

3. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 2, characterized in that: The value range of β mentioned in step S21 is [0.1, 0.9]. In the knowledge graph experiment based on administrative divisions, the ARI value is optimal when β = 0.6, which is 0.75; and in this experiment, the clustering stability is highest when γ = 0.

25. The range of θ in step S23 is [0.1, 1], with the optimal value being 0.

5.

4. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 2, characterized in that: The hierarchical threshold mentioned in step S32 Through user interaction testing, it was determined that the information density of the visual interface is moderate at different zoom levels, without being overly crowded or overly sparse. The graph embedding method described in step S12 is an unsupervised embedding algorithm based on graph structure, including Node2Vec, DeepWalk, or GCN, to effectively capture the topological structure information of the knowledge graph.

5. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 1, characterized in that: The increment δ of the number of sub-clusters in step S23 is a positive integer, and its specific value is adaptively adjusted according to the size of the cluster to ensure that each sub-cluster contains a sufficient number of nodes to maintain clustering stability.

6. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 2, characterized in that: The backend database mentioned in step S31 uses a database system that supports spatiotemporal indexes, including PostgreSQL with PostGIS extensions, to improve the node subset. Query efficiency.

7. The hierarchical progressive visualization method for nebula diagrams of large-scale spatiotemporal knowledge graphs according to claim 1, characterized in that: The initial number of clusters K mentioned in S22 is determined by combining the elbow rule with the clustering effect evaluation to ensure that the clustering results avoid both fragmentation caused by too many clusters and excessive information aggregation caused by too few clusters.

Images