Graph visualization controllable layout interactive generation method based on depth extension model

Abstract

The invention discloses a graph visualization controllable layout interactive generation method based on a depth extension model, and belongs to the field of visualization and human-computer interaction. Comprising the following steps: (1) graph data enhancement; (2) designing a deep diffusion model structure oriented to graph layout; (3) pre-training a deep diffusion model oriented to graph layout; (4) the user uploads graph data in a JSON format; (5) constructing an expected layout constraint through user interaction; (6) the system analyzes the layout constraint and generates a graph visualization layout based on the pre-training model; (7) visualizing a layout result by the system display graph and supporting the user to explore the output of different models; and (8) a user iteratively creates layout constraints to construct a data stream type visual exploration process. According to the method, the graph layout method based on the deep diffusion model and the interactive graph visualization generation system are provided, an analyst without a programming background is supported to efficiently construct a graph visualization result meeting user requirements, and the difficulty of screening a proper graph visualization layout form by a user is reduced.

Description

Technical Field

[0001] The present invention relates to the field of visualization and human-computer interaction, and specifically proposes a graph visualization controllable layout interactive generation method based on a deep extension model for the node-link graph visualization form that is commonly found in practical applications. Background Art

[0002] Graph data is widely used in various fields of practical applications, including social networks, molecular structures, and citation networks. Graph visualization supports users to intuitively understand the internal connections and topological structures of data. Among various methods of graph visualization, node-link graphs are the most widely used, which supports users to understand the topological structure and connection relationships of graph data. In this case, data objects are mapped to nodes, and the connection relationships between data objects are represented as lines between nodes. At present, many algorithms have been proposed in the field of visualization research to layout node-link graphs. Different algorithms are suitable for different graph data types and analysis tasks. The layout form of the same graph is diverse depending on the selected algorithm and its specific configuration.

[0003] In the process of determining the visualization form of the node-link graph, users often need to combine their own preferences and their understanding of the data to select the most suitable visualization form for their needs, thereby assisting users in analyzing graph data. Therefore, it is a very important task to support users in efficiently obtaining a visualization layout that meets their needs. At present, the main way to obtain a graph layout that meets user needs is to write code and interactively select templates. However, the method of writing code supports users to explore and obtain a layout that meets their expectations by adjusting the layout algorithm parameters. This method is difficult to learn and has a heavy programming burden, and is not suitable for most non-expert users. Although the interactive template selection method can improve the efficiency of user layout selection, this process requires users to traverse all graph layout algorithms, which is cumbersome and time-consuming, and cannot fully utilize users' expectations for graph layout results. At present, there is no method that can fully meet users' expectations and efficiently and accurately generate graph visualization layout forms. Summary of the invention

[0004] In view of the defects existing in the prior art, the purpose of the present invention is to provide a method for interactively generating a graph visualization controllable layout based on a deep diffusion model for node link graph data. The method supports users to efficiently obtain graph visualization results that meet their expectations. Users specify the positions of specific nodes as constraints of the graph layout based on their expectations for the graph layout, and these nodes can form a connected subgraph or multiple unconnected parts. The system includes two parts, the front end and the back end. Users load the graph data to be processed through the system front end, and users can add layout constraints to this data, and then the system front end transmits these data and possible constraints to the system back end. The system back end parses the layout constraints and generates a graph visualization layout based on a pre-trained model, and transmits the generated data to the front end, and the front end displays the graph visualization layout results based on these data. The entire system automatically generates a graph visualization layout form according to the constraints provided by the user, and ensures that the generated layout results match the user's constraints and the diversity of the output graph layout results, thereby directly generating a graph visualization result that meets the user's expectations, reducing the difficulty for users to explore different graph visualization forms.

[0005] To achieve the above purpose, the technical solution adopted by the present invention for generating a visualized node link graph is:

[0006] An interactive generation method for controllable graph visualization layout based on a deep extension model includes the following eight steps: (1) graph data enhancement; (2) graph layout model structure design; (3) graph layout model pre-training; (4) users upload graph data in JSON format through the system front end; (5) users construct desired layout constraints through the system front end interaction; (6) the system back end parses the layout constraints and generates a graph visualization layout based on the pre-trained model; (7) the system front end displays the graph visualization layout results and supports users to explore the outputs of different models; (8) users iteratively create layout constraints to build a data flow visual exploration process.

[0007] Specific:

[0008] (1) Graph data enhancement;

[0009] Graph data enhancement involves the enhancement of node features and the enhancement of graph structure. Among them, the enhancement of node features is crucial for generating the layout of non-attribute graphs. Since the graph layout depends largely on the topological structure of the graph, the choice of node encoding method must accurately represent the topological features of the graph. Therefore, the present invention adopts the Laplace position encoding method as the initial node feature. Laplace position encoding is a spectral technology that embeds the topological graph into Euclidean space. The node encoding result combines position and structural information to establish a meaningful local coordinate system while retaining the global graph structure. Nodes far away in the graph have different encodings, while close nodes should have similar encodings. Laplace position encoding L is defined by factoring the graph Laplace matrix:

[0010]

[0011] Here I is the N×N identity matrix, D is the node degree matrix, A represents the adjacency matrix, Λ and U represent the eigenvalue and eigenvector respectively. The present invention uses the k smallest non-trivial eigenvectors as the initial node features, and k is a hyperparameter. The eigenvector encodes the smooth position coordinates of the adjacent nodes, so that the nodes can effectively capture the structural information in their perception domain during the message transmission process, thereby helping the model to better obtain the topological relationship between nodes.

[0012] In addition, the present invention enhances the original graph topology to improve the efficiency of message transmission between nodes. Graph structure enhancement involves establishing connections between nodes, which are divided into three types: original edges, local edges and global edges.

[0013] (a) Original edges represent edges from the original graph, preserving the inherent connections in the given graph and directly reflecting its topological structure.

[0014] (b) Local edges connect each node to its n-th order neighbors as an extension of the original graph topology, which expands the perception range of the node and helps the model effectively generate the local structure around each node, where n is determined by hyperparameter tuning of grid search.

[0015] (c) Both the original edge and the local edge are derived from the topological structure of the original graph, which promotes the information exchange between topologically adjacent nodes. However, in the graph layout, each node is in a different position in the two-dimensional space. Nodes with close physical positions can significantly affect the graph layout, but they may have a considerable topological distance, which limits their information exchange. Therefore, the present invention introduces global edges for dynamically connecting nodes within a distance threshold τ, and the radius cutoff value is determined according to the specific scale of the layout. Global edges contribute to a more consistent and comprehensive global structure.

[0016] (2) Structural design of deep diffusion model for graph layout

[0017] Based on the diffusion model, a node-link graph generation model is designed, which mainly includes two processes: a forward process and a reverse process. The present invention defines a graph layout as a system consisting of multiple nodes and edges in a two-dimensional space. The forward process involves transforming the system from an equilibrium state to a chaotic state, while the reverse process optimizes the system from the chaotic state to restore it to an equilibrium state. The equilibrium state represents an orderly layout, while the chaotic state represents a chaotic and noisy layout. The present invention defines X0 as the graph layout in an equilibrium state. X t The overall goal of the model is to learn how to go from X T For a given graph G, the forward process gradually introduces slight Gaussian noise (sampled from a standard Gaussian distribution) into the layout X0. After several steps of perturbation, it is transformed into a completely noisy layout X T , while ensuring that X T follows the prior distribution N(0,1). At each step, the signal-to-noise ratio is determined by a fixed variance pattern β1,…,β T The present invention represents the forward process as a Markov chain:

[0018]

[0019] That is, for a given graph G, the state Xt, under the condition of the previous state Xt-1, follows a Gaussian distribution in represents the mean of the Gaussian distribution, β t I represents the covariance matrix, where I is the identity matrix;

[0020] For t independent and identically distributed Gaussian noises N(μ, σ 2 ), which is still a Gaussian noise after superposition; in order to simplify this process, the present invention defines α t =1-β t and This enables the present invention to sample X given X0 at any time step t. t :

[0021]

[0022] That is, for a given graph G, state X t Under the condition of initial state X0, it obeys a Gaussian distribution This means that for any state X of the graph G t , you can add the corresponding Gaussian noise to the initial state X0 get;

[0023] The inverse process is an iterative optimization process that samples a completely noisy layout X from the prior distribution N(0, 1)T ; This process involves T denoising steps, gradually optimizing the layout system from a chaotic state to a balanced state. The reverse process is also a Markov chain:

[0024]

[0025] That is, for a given graph G, state X t-1 In state X t Under the condition of

[0026] Among them, μ θ is the function used to estimate the mean of the distribution, σ t 2 represents an arbitrary variance. μ θ The calculation method is as follows:

[0027]

[0028] ∈ θ is a trainable function used to predict the layout X t The noise on is also the training target of the model of the present invention:

[0029]

[0030] This function is a loss function used to measure the noise ∈ predicted at the tth step θ The difference between the actual noise ∈ and the actual noise ∈; By minimizing this loss function, we can train an algorithm to accurately predict future noise ∈ θ Model;

[0031] For the graph layout generation task, it is crucial to ensure that the graph layout is invariant under rotation and translation. Therefore, the function ∈ θ The invariance of rotation and translation must be guaranteed. The present invention uses an equivariant graph neural network to parameterize ∈ θ The equivariant graph neural network consists of multiple layers of equivariant convolutional layers:

[0032]

[0033] The first formula represents the message passing mechanism in equivariant graph neural networks: each node i sends a message m to its neighbor node j. ij The content of the message includes the hidden states of nodes i and j in the current layer. and The square of the distance between them And the edge feature e ij φe is a fully connected neural network that aggregates edge endpoint embeddings and node distances;

[0034] The second formula represents the mechanism by which the equivariant graph neural network updates the hidden state of a node: the digital state of the next layer of node i It is obtained by summing up the messages from all neighbors and then passing them through a fully connected neural network φ h Updated. Here N(i) represents the neighbor set of node i;

[0035] The third formula represents the mechanism by which the equivariant graph neural network updates the coordinates of nodes: the digital coordinates of the next layer of node i are Through the current coordinates It is obtained by adding a correction term. The correction term takes into account the relative position information between nodes and the updated hidden state. x It is a fully connected neural network responsible for projecting the updated endpoint embeddings and edge embeddings into two-dimensional space;

[0036] Where: l is the index of the layer; h l represents the node embedding vector, and the initial value is obtained from the Laplace eigenvector position encoding; x l Corresponding to the node coordinates, quantifies the Euclidean distance between node i and node j; e ij represents the type of edge, including primitive, local, and global edges; N(i) represents the neighborhood of node i; φe is a fully connected neural network used to aggregate edge endpoint embeddings and node distances; φ h The node embedding is updated by utilizing the information of neighboring nodes and parameterized using a graph isomorphic neural network; since the node coordinates are weighted by the invariant feature x i -x j Update,∈ θ The rotation and translation invariance of the graph layout is ensured.

[0037] In addition, in order to reduce the layout burden of the model, the present invention ensures that the position of the overall layout remains relatively stable in the two-dimensional space, and transforms the centroid of the X0 layout to the origin to obtain the layout The centroid represents the weighted average of all node positions in the layout. The specific transformation operation is as follows:

[0038]

[0039] (3) Pre-training of deep diffusion models for graph layout

[0040] For the pre-training of the graph layout model, the present invention collects the layout results of several graphs, including different types of graphs and the layout results of the same graph using different layout algorithms. During the training process, the present invention selects a random time step t and adds noise ∈ sampled from the prior distribution N(0, 1) to the equilibrium graph layout X0. The signal-to-noise ratio is determined by t and the predefined variance patterns β1, ..., β T Determine, perturb and get the chaotic layout X t Next, the disturbed layout X t Input to the denoising network ∈ θ and use ∈ as a supervisory signal to train the network’s ability to predict layout noise. θ Can accurately predict X t When there is noise on the graph, the model can restore the chaotic graph layout to a high-quality graph layout in equilibrium through the inverse process defined above.

[0041] (4) Users upload graph data in JSON format through the system front end;

[0042] Users load and upload graph data in JSON format to the system backend through the system frontend. The system backend automatically parses the data after receiving it, and enhances the original graph data, enriches the node attributes in the graph, and enhances the topological structure of the graph. The system will use the pre-trained model to generate an initial layout without any constraints for the uploaded graph data. The output results of the model will be automatically visualized in the system.

[0043] (5) The user constructs the desired layout constraints through the system front-end interaction;

[0044] This system supports users to quickly build layout constraints according to their expectations for graph layout to guide the generation of graph layout. Users select specific nodes in the system front-end visualization results and move them to their desired locations, or move multiple nodes collectively by selecting and dragging. The system control panel provides the user with the expected layout constraint information added for this graph. In order to help users efficiently build graph layout constraints, the system front-end provides four common interactive operations: horizontal flip, vertical flip, zoom, rotation, and four commonly used layout modes to arrange the selected nodes into corresponding shapes: straight line, spiral, circle and regular polygon. By building layout constraints, users can control the graph layout generation process and guide the model to generate layout results that are compatible with the layout constraints, thereby generating a graph layout visualization that meets user expectations.

[0045] (6) The system backend parses the layout constraints and generates a graph visualization layout based on the pre-trained model;

[0046] The system backend parses the layout constraints constructed by the user in step (5) into acceptable inputs for the model and integrates them into each step of the layout generation process. At each time step t in the reverse process in step (2), the present invention merges the constraints provided by the user with the generation results of the model. Considering the rotation and translation invariance of the graphic layout, there is usually a deviation in the overall position of the layout constraints and the model generation results in two-dimensional space. In order to promote effective coordination between the layout constraints and the model generation results, the present invention performs resampling at each time step t, by X t-1 Back to X t ,This process is repeated n times to ensure that the model can align the layout constraints with the generated results. Through the constrained generation process, the model can seamlessly integrate the user layout constraints into the model generation results.

[0047] (7) The system front end displays the visualization layout results and supports users to explore the outputs of different models;

[0048] The system front-end visualization module visualizes the graph layout results generated by the system back-end model. The visualization part of the system is completed based on the visualization toolkit D3. The graph visualization drawing results can be automatically adjusted according to the window size. After the graph visualization is displayed on the computer screen, the user can interact with the graph visualization, rotate, zoom or interactively select a part of the data for detailed viewing, so as to explore the graph layout results. At the same time, users can select pre-trained models obtained from different data sets and adjust the layout hyperparameters to explore the graph layout results of different models.

[0049] (8) Users iteratively create layout constraints to build a data flow visual exploration process;

[0050] The system supports users to iteratively create layout constraints to build a data flow visual exploration process. The system draws on the classic data flow model. Users define layout constraints in each module of the data flow according to step (5). The system generates a graph layout result based on the provided constraints, and the layout result serves as the input of the next module. When the data in a module is updated, the input and output of all subsequent modules connected to it will be automatically updated. At the same time, in order to support the reuse of graph layout constraints, the system also provides a module copy function. Users can copy some modules in the data flow, and the input of the copied sub-data flow is determined by the new connection relationship. Users can iteratively create layout constraints to adjust the graph layout results until they get a graph layout visualization form that meets their needs.

[0051] The beneficial effects of the present invention are:

[0052] The method provided by the present invention supports users to interactively construct layout constraints to guide the graph visualization generation process, thereby directly obtaining a graph visualization layout form that meets the user's expectations. Based on the pre-trained graph layout model, the system introduces an interactive layout constraint construction method and a data flow model to realize a complete exploratory graph visualization form generation system, which supports users to efficiently generate graph visualization forms according to their needs.

[0053] Specifically, the effects of the present invention include:

[0054] Capable of generating node-link graph visualization: In the above steps (2) and (3), a graph visualization generation model based on a diffusion model is designed and pre-trained to support the generation of a suitable node-link visualization layout for graph data.

[0055] Supporting the user to guide the graph layout generation process and reducing the difficulty of obtaining the desired layout: the above step (5) supports the user to construct the layout constraints that reflect their expectations, and step (6) generates a layout result that matches the layout constraints provided by the user. The generation process does not require the user to screen the layout results and does not require the user to have a programming foundation, so the method has a lower learning curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 For the operation flow of users in the implementation mode of the present invention;

[0057] Figure 2 It is a schematic diagram of the topological structure enhancement in a specific implementation manner of the present invention;

[0058] Figure 3 A schematic diagram of a flow chart of generating a layout constraint in a specific implementation manner of the present invention;

[0059] Figure 4 A graphical interface diagram is constructed for interactive exploration of graph visualization and layout constraints in a specific embodiment of the present invention;

[0060] Figure 5 A comparison diagram of the layout generation result and the reference layout in a specific implementation manner of the present invention;

[0061] Figure 6 A comparison diagram of the tree diagram generation result and the reference layout in a specific implementation manner of the present invention;

[0062] Figure 7 A comparison diagram of the generation results of multiple types of graph data and the reference layout in a specific implementation manner of the present invention;

[0063] Figure 8 It is a schematic diagram of constraint generation effect in a specific implementation manner of the present invention. DETAILED DESCRIPTION

[0064] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0065] Figure 1 The user's operation flow in the embodiment of the present invention is shown.

[0066] Figure 2 The figure shows the way of enhancing the graph topology in a specific embodiment of the present invention, taking the yellow node as an example. The figure on the left highlights the original edge connected to it. The figure in the middle highlights the second-order local edge, and the figure on the right highlights the global face change, which is connected according to the predefined distance threshold τ.

[0067] Figure 3 The constraint generation process of the graph layout in the specific embodiment of the present invention is shown to sample a layout X with noise. T ~N(0,1), and then merge it with the constraints derived from the user's layout expectations. After relocating the center of mass of the layout, the present invention obtains a tensor X representing the node position. Further through graph enhancement, tensors E and A representing edge types and adjacency matrices of the graph are generated. The present invention uses Laplace eigenvectors for position encoding to obtain a tensor H representing the initial node embedding. The above tensor is input into the noise predictor for denoising. In order to effectively coordinate the given constraints and model generation, each time step undergoes n resamplings. The complete generation process iterates T times, and finally generates X0 as the output of the model.

[0068] Figure 4 The system front-end graph visualization exploration and layout constraint interactive construction interface in a specific embodiment of the present invention is shown. Each module of the exploration data flow corresponds to a layout constraint and graph data, and the model generation result will be directly displayed in the form of graph visualization. Figure 4 ,The system diagram visualization panel is used to visualize the generation results of the ,backend model. The yellow nodes in the diagram are the ,constraint information added by the user.

[0069] Figure 5 The figure shows the comparison between the graph layout generation result and the reference layout in the specific implementation manner of the present invention.

[0070] Figure 6 The comparison between the generation effect of the tree map in a specific embodiment of the present invention and the reference layout is shown, wherein the model used is trained on a benchmark map dataset, which illustrates the generalization ability of the model.

[0071] Figure 7 A comparison between various different types of graph layouts generated in a specific embodiment of the present invention and a reference layout is shown. All graph layouts are generated by one model, which verifies the model's ability to simultaneously learn different types of graphs and their corresponding visualization layout results.

[0072] Figure 8 The layout effect obtained by using layout constraints in a specific embodiment of the present invention is shown, and the yellow nodes represent the layout constraints provided to the model. The two layouts in each row depict the same graph, and all graph layouts are generated by one model, which verifies the ability of model constraint generation.

Claims

1. A graph visualization controllable layout interactive generation method based on a deep extension model, characterized in that: The method includes the following eight steps: (1) graph data enhancement; (2) graph layout model structure design; (3) graph layout model pre-training; (4) users upload graph data in JSON format through the system front end; (5) users build the desired layout constraints through the system front end interaction; (6) the system back end parses the layout constraints and generates a graph visualization layout based on the pre-trained model; (7) the system front end displays the graph visualization layout results and supports users to explore the outputs of different models; (8) users iteratively create layout constraints to build a data flow visual exploration process.

2. The interactive generation method of graph visualization controllable layout based on the deep extension model according to claim 1 is characterized in that: The specific method of step (1) is: Graph data enhancement involves the enhancement of node features and graph structure; the Laplace position encoding method is used as the initial node feature; The Laplacian positional encoding L is defined by factorizing the graph Laplacian matrix: Here, is the N×N identity matrix, D is the node degree matrix, A represents the adjacency matrix, Λ and U represent the eigenvalue and eigenvector respectively; the k smallest non-trivial eigenvectors are used as the initial node features, and k is a hyperparameter; The original graph topology is enhanced to improve the efficiency of message transmission between nodes. Graph structure enhancement involves establishing connections between nodes, which are divided into three types: original edges, local edges, and global edges. (a) Original edges represent edges from the original graph, which preserve the inherent connections in the given graph and directly reflect its topological structure; (b) Local edges connect each node to its n-th order neighbors as an extension of the original graph topology, which expands the node’s perception range and helps the model effectively generate the local structure around each node, where n is determined by hyperparameter adjustment of grid search; (c) Global edges are used to dynamically connect nodes within a distance threshold τ, and the radius cutoff is determined according to the specific scale of the layout.

3. The interactive generation method of graph visualization controllable layout based on the deep extension model according to claim 1 is characterized in that: The specific method of step (2) is: The generation model for the node-link graph includes two processes: the forward process and the reverse process; A graph layout is defined as a system of nodes and edges in two-dimensional space; the forward process involves transforming this system from an equilibrium state to a chaotic state, while the reverse process optimizes the system from the chaotic state to restore it to an equilibrium state; The equilibrium state represents an orderly layout, while the disordered state represents a chaotic and noisy layout; define X o is the layout of the equilibrium state; X t The overall goal of the model is to learn how to go from X to T Recovery X o ; For a given graph G, the forward process gradually introduces slight Gaussian noise to the layout X o In; through several perturbations, it is transformed into a completely noisy layout X T , while ensuring that X T follows the prior distribution N(0,1); at each step, the signal-to-noise ratio is determined by a fixed variance pattern β1,…,β T Decision; represent the forward process as a Markov chain: That is, for a given graph G, state X t In the previous state X t-1 Under the condition of in represents the mean of the Gaussian distribution, β t I represents the covariance matrix, where is the identity matrix; For t independent and identically distributed Gaussian noises N(μ, σ 2 ), which is still a Gaussian noise after superposition; therefore, define α t =1-β t and Then we can give X at any time step t o Sampling X t : That is, for a given graph G, state X t Under the condition of initial state X0, it obeys a Gaussian distribution This means that for any state X of the graph G t , you can add the corresponding Gaussian noise to the initial state X0 get; The inverse process is an iterative optimization process that samples a completely noisy layout X from the prior distribution N(0, 1) T ; This process involves T denoising steps, gradually optimizing the layout system from a chaotic state to a balanced state; the reverse process is also a Markov chain: That is, for a given X t-1 The graph G, the state is in state X t Under the condition of Among them, μ θ is a function used to estimate the mean of a distribution, represents arbitrary variance; μ θ The calculation method is as follows: ∈ θ is a trainable function used to predict the layout X t The noise on is also the training target: This function is a loss function used to measure the noise ∈ predicted at the tth step θ The difference between the actual noise ∈ and the actual noise ∈; By minimizing this loss function, we can train a neural network to accurately predict future noise ∈ θ Model; We use equivariant graph neural network to parameterize ∈ θ ; The equivariant graph neural network consists of multiple layers of equivariant convolutional layers: The first formula represents the message passing mechanism in equivariant graph neural networks: each node i sends a message m to its neighbor node j. ij ; The content of the message includes the hidden state of nodes i and j in the current layer and The square of the distance between them And the edge feature e ij ; φ e is a fully connected neural network that aggregates edge endpoint embeddings and node distances; The second formula represents the mechanism by which the equivariant graph neural network updates the hidden state of a node: the digital state of the next layer of node i It is obtained by summing up the messages from all neighbors and then passing them through a fully connected neural network φ h Updated; here N(i) represents the neighbor set of node i; The third formula represents the mechanism by which the equivariant graph neural network updates the coordinates of nodes: the digital coordinates of the next layer of node i are Through the current coordinates Add a correction term to get; the correction term takes into account the relative position information between nodes and the updated hidden state; φ x It is a fully connected neural network responsible for projecting the updated endpoint embeddings and edge embeddings into two-dimensional space; Where: l is the index of the layer; h l represents the node embedding vector, and the initial value is obtained from the Laplace eigenvector position encoding; x l Corresponding to the node coordinates, quantifies the Euclidean distance between node i and node j; e ij represents the type of edge, including primitive, local, and global edges; N(i) represents the neighborhood of node i; φe is a fully connected neural network used to aggregate edge endpoint embeddings and node distances; φ h The node embedding is updated by utilizing the information of neighboring nodes and parameterized using a graph isomorphic neural network; since the node coordinates are weighted by the invariant feature x i -x j Update,∈ θ Ensures rotation and translation invariance of graph layout; X o The centroid of the layout is transformed to the origin to obtain the graph layout The centroid represents the weighted average of all node positions in the layout. The specific transformation operation is as follows:

4. The interactive generation method of graph visualization controllable layout based on the deep extension model according to claim 1 is characterized in that: The specific method of step (3) is: For the pre-training of the graph layout model, we collect the layout results of several graphs, including different types of graphs and the layout results of the same graph using different layout algorithms. During the training process, we select a random time step t and add noise ∈ sampled from the prior distribution N(0, 1) to the equilibrium graph layout X0. The signal-to-noise ratio is determined by t and the predefined variance patterns β1, …, β T Determine, perturb and get the chaotic layout X t ; Then, the disturbed layout X t Input to the denoising network ∈ θ and use ∈ as a supervisory signal to train the network's ability to predict layout noise; when ∈ θ Can accurately predict X t When there is noise on the graph, the model restores the chaotic graph layout to a high-quality graph layout in the equilibrium state through the defined inverse process.

5. The interactive generation method of graph visualization controllable layout based on deep extension model according to claim 1 is characterized in that: The specific method of step (4) is: The user loads and uploads graph data in JSON format to the system backend through the system frontend. After receiving the data, the system backend automatically parses the data and enhances the original graph data, enriches the node attributes in the graph, and enhances the topological structure of the graph. The system will use the pre-trained model to generate an initial layout without any constraints for the uploaded graph data. The output of the model is automatically visualized in the system.

6. The interactive generation method of graph visualization controllable layout based on deep extension model according to claim 1 is characterized in that: The specific method of step (5) is: It supports users to quickly build layout constraints according to their expectations for graph layout to guide the generation of graph layout. Users select specific nodes in the system front-end visualization results and move them to their desired positions, or move multiple nodes collectively by selecting and dragging. The system control panel provides users with the expected layout constraint information added for this graph. In order to help users efficiently build graph layout constraints, the system front-end provides four common interactive operations: horizontal flip, vertical flip, zoom, rotation, and four commonly used layout modes to arrange the selected nodes into corresponding shapes: straight line, spiral, circle and regular polygon. By building layout constraints, users control the graph layout generation process and guide the model to generate layout results that are compatible with the layout constraints, thereby generating a graph layout visualization form that meets user expectations.

7. The interactive generation method of graph visualization controllable layout based on deep extension model according to claim 1 is characterized in that: The specific method of step (6) is: The system backend parses the layout constraints constructed by the user in step (5) into acceptable inputs for the model and integrates them into each step of the layout generation process; at each time step t in the reverse process in step (2), the constraints provided by the user are merged with the generation results of the model; considering the rotation and translation invariance of the graphic layout, there is a deviation in the overall position of the layout constraints and the model generation results in two-dimensional space; in order to promote effective coordination between the layout constraints and the model generation results, resampling is performed at each time step t, through X t-1 Back to X t , this process is repeated n times to ensure that the model can align the layout constraints and the generated results; through the constrained generation process, the model can seamlessly integrate the user layout constraints into the model generation results.

8. The interactive generation method of graph visualization controllable layout based on deep extension model according to claim 1 is characterized in that: The specific method of step (7) is: The system front-end visualization module visualizes the graph layout results generated by the system back-end model; the visualization part of the system is completed based on the visualization toolkit D3; the graph visualization form drawing results are automatically adjusted according to the window size; After the graph visualization is displayed on the computer screen, the user interacts with the graph visualization, rotates, zooms, or interactively selects a portion of the data for detailed viewing, thereby exploring the graph layout results; at the same time, the user selects pre-trained models obtained from different data sets and adjusts layout hyperparameters to explore the graph layout results of different models.

9. The interactive generation method of graph visualization controllable layout based on deep extension model according to claim 1 is characterized in that: The specific method of step (8) is: The system supports users to iteratively create layout constraints to build a data flow visual exploration process; the system draws on the classic data flow model, and users define layout constraints in each module of the data flow according to step (5). The system generates a graph layout result based on the provided constraints, and the layout result serves as the input of the next module; when the data in a module is updated, the input and output of all subsequent modules connected to it will be automatically updated; at the same time, in order to support the reuse of graph layout constraints, the system also provides a module copy function, and users can copy some modules in the data flow. The input of the copied sub-data flow is determined by the new connection relationship; users can iteratively create layout constraints to adjust the graph layout results until they get a graph layout visualization form that meets their needs.

Images