Method for carrying out multi-business data visualization generation by using AI intelligent agent
By using AI intelligent agents to generate multi-business data visualizations, and utilizing feature extraction and graph neural network processing, the semantic association problem of multi-source heterogeneous data is solved, generating clear visualization graphics that can adapt to the needs of multiple fields and achieve efficient data integration and visualization effects.
Patent Information
- Application Number
- CN202511463664.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-14
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional single data collection and analysis methods are insufficient to meet the comprehensive decision-making needs of multi-dimensional and cross-domain applications. In particular, in scenarios where multiple business formats such as retail, finance, manufacturing, logistics, healthcare, and urban governance are developing in parallel, multi-source heterogeneous data exhibit diverse, dynamic, and heterogeneous characteristics. Furthermore, there is insufficient semantic extraction and representation of high-dimensional data, low automation of semantic association mining, and poor visualization of high-dimensional data.
The AI intelligent agent generates multi-business data visualizations, including feature extraction, semantic association graph clustering, graph neural network processing, and uniform manifold approximation projection algorithm. It generates semantic clusters and performs dimensionality reduction and graphics rendering. It uses similarity graphs and graph neural networks to capture deep semantic associations, dynamically adjusts similarity weights, and generates clear visualization graphics.
It achieves effective integration and unified semantic expression of data from multiple business sectors, reduces data processing complexity, improves the intuitiveness and cross-domain applicability of visualization results, and ensures a high degree of matching between visualization results and specific scenarios.
Smart Images
Figure CN120950752A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence, and in particular to a method for generating multi-business data visualizations using AI agents. Background Technology
[0002] In the rapid development of the digital economy and intelligent society, data has become a core resource driving innovation and development for enterprises, governments, and society. With the widespread adoption of multi-industry integrated business models, traditional single-data collection and analysis methods are no longer sufficient to meet the comprehensive decision-making needs across multiple dimensions and fields, especially in scenarios where multiple industries such as retail, finance, manufacturing, logistics, healthcare, and urban governance are developing in parallel, resulting in diverse, dynamic, and heterogeneous data sources. How to effectively integrate multi-industry data through intelligent means and present it intuitively in a visual manner has become an important direction for current technological research and industrial application. The method of using AI agents to generate multi-industry data visualizations is a crucial technological direction for addressing the needs of multi-source heterogeneous data fusion and dynamic intelligent decision-making. It not only breaks through the limitations of traditional data visualization methods but also achieves cross-industry and cross-modal visualization generation through the autonomous learning and multimodal processing capabilities of intelligent agents. With the continuous advancement of data infrastructure, artificial intelligence algorithms, and interactive technologies, it is expected to be widely applied in multiple fields such as fintech, smart retail, intelligent manufacturing, healthcare, and smart cities in the future, promoting the construction of a data-driven intelligent society. Summary of the Invention
[0003] This invention provides a method for generating multi-business data visualizations using AI agents, in order to solve the technical problems of difficulty in handling heterogeneous multi-business data, insufficient semantic extraction and representation of high-dimensional data, low degree of automation in semantic association mining, and poor visualization effect of high-dimensional data.
[0004] The present invention provides a method for generating multi-business data visualization using AI agents, specifically comprising the following technical solutions: A method for generating multi-business data visualizations using AI agents includes the following steps: S1. Obtain data from various data sources and extract features to obtain feature vectors. Integrate all feature vectors into a data base and extract semantic relationships from the data base using a semantic association graph clustering algorithm to generate semantic clusters. S2. Based on semantic clusters and data foundation, cluster enhancement, dimensionality reduction, and graphic rendering are performed through multi-business data visualization generation algorithms to generate visualized graphics of multi-business data.
[0005] Preferably, S1 specifically includes: In the implementation of the semantic association graph clustering algorithm, cosine similarity is calculated for each pair of feature vectors in the data base; cosine similarity is calculated by dividing the inner product of feature vectors by the product of their moduli, generating edge weights, i.e., similarity weights.
[0006] Preferably, S1 specifically includes: A similarity graph is constructed based on similarity weights. Nodes in the similarity graph represent feature vectors, and edges connect each pair of nodes. The edge weight is the corresponding cosine similarity.
[0007] Preferably, S1 specifically includes: A graph neural network is used to perform multi-layer iterative processing on the similarity graph, and the final enhanced feature representation of the nodes is output. Based on the final enhanced feature representation of the nodes, a semantic cluster is output.
[0008] Preferably, S2 specifically includes: Based on the cosine similarity between feature vectors in the data base, the cosine similarity between all feature vectors within each semantic cluster is averaged to obtain the intra-cluster cohesion.
[0009] Preferably, S2 specifically includes: For each pair of semantic clusters, iterate through all feature vectors of one cluster and all feature vectors of the other cluster, calculate the cosine similarity, and take the minimum value among all cosine similarities as the inter-cluster separation degree.
[0010] Preferably, S2 specifically includes: Based on intra-cluster cohesion and inter-cluster separation, the similarity weights between feature vectors are dynamically adjusted; for feature vector pairs within the same cluster, the cosine similarity is multiplied by an enhancement factor; for feature vector pairs from different clusters, the cosine similarity is multiplied by a weakening factor to obtain the adjusted similarity weights.
[0011] Preferably, S2 specifically includes: Based on the adjusted similarity weights, a uniform manifold approximation and projection algorithm is used to map high-dimensional feature vectors to a low-dimensional space to generate visual coordinates. In the low-dimensional space, initial low-dimensional coordinates are randomly generated for each feature vector as the starting point for the optimization of the uniform manifold approximation and projection algorithm. The low-dimensional coordinates are optimized by minimizing the difference between the high-dimensional similarity and the low-dimensional similarity distribution.
[0012] The beneficial effects of the technical solution of the present invention are: 1. By employing principal component analysis, pre-trained convolutional neural networks, and pre-trained language models to extract features from structured, semi-structured, and unstructured data respectively, the core semantic information of data from each data source was successfully captured, generating a unified data foundation. This effectively solved the problem of strong heterogeneity among multiple data sources, ensuring that the data has a consistent semantic expression in the high-dimensional vector space, providing a unified semantic foundation for subsequent analysis, and reducing the complexity of data processing.
[0013] 2. By constructing a similarity graph and combining it with a graph neural network, we can capture the deep semantic relationships between feature vectors. The similarity graph measures the semantic proximity between feature vectors based on cosine similarity, while the graph neural network enhances the node feature representation through multi-layer iteration, reflecting the global and local relationships of feature vectors in the semantic space. This significantly improves the discriminative power of the enhanced feature representation, automatically generates semantic clusters without manual intervention, and reduces the complexity of multi-business data analysis.
[0014] 3. By calculating intra-cluster cohesion and inter-cluster separation, the similarity weight is dynamically adjusted, which effectively strengthens the compactness of intra-cluster feature vectors and highlights inter-cluster differences. By combining uniform manifold approximation and projection algorithms, high-dimensional feature vectors are mapped to low-dimensional space to generate visual coordinates that preserve semantic relationships. This ensures that the visualization graphics retain the high-dimensional semantic structure while having clear intra-cluster aggregation and inter-cluster separation effects, thus improving the intuitiveness and readability of the visualization results.
[0015] 4. By automatically selecting suitable visualization formats, such as scatter plots, relationship diagrams, and heatmaps, from a predefined template library based on the application scenario, and combining them with mature graphics libraries to generate graphics, it can flexibly adapt to the needs of different fields. This method ensures that the visualization results are highly matched with the specific scenario, improving the universality and practicality of cross-domain applications. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a method for generating multi-business data visualization using AI agents, as described in this invention. Detailed Implementation
[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for a method of generating multi-business data visualization using AI intelligent agents provided by the present invention.
[0020] See attached document Figure 1 The diagram illustrates a flowchart of a method for generating multi-business data visualization using an AI agent, according to an embodiment of the present invention. The method includes the following steps: S1. Obtain data from various data sources and extract features to obtain feature vectors. Integrate all feature vectors into a data base and extract semantic associations from the data base using a semantic association graph clustering algorithm to generate semantic clusters. This approach acquires data from multiple data sources, the specific types of which can be determined based on the application scenario. For example, in the industrial field, this might involve equipment operation logs and equipment maintenance reports. These multiple data sources include three data types: structured data, semi-structured data, and unstructured data, such as tables, images, and text. For each data type, a specific feature extraction method is selected to capture the core semantic information of each data source. Specifically, principal component analysis is used for structured data, pre-trained convolutional neural networks are used for semi-structured data, and pre-trained language models, such as natural language processing models, are used for unstructured data. Feature vectors are extracted for each data source, and then all feature vectors from all data sources are integrated into a unified data foundation, represented as a set containing all feature vectors. , The number of feature vectors is indicated; the feature extraction methods described all employ techniques well-known in the art and will not be elaborated upon here.
[0021] By selecting specific feature extraction methods for different data types, the problem of strong heterogeneity among multiple data sources can be solved, ensuring that the data has a consistent semantic expression in a high-dimensional vector space.
[0022] Semantic clusters are generated by extracting semantic associations from the data base using a semantic association graph clustering algorithm. This algorithm measures the semantic closeness between each pair of feature vectors by calculating cosine similarity. The cosine similarity is calculated by dividing the inner product of the feature vectors by their modulo product, generating a value... The edge weights between two feature vectors, also known as similarity weights, represent the correlation between them in the semantic space. Based on these similarity weights, an undirected graph is constructed, called the similarity graph. In the similarity graph, each node represents a feature vector, and the edges connecting each pair of nodes have weights that represent the corresponding cosine similarity. The calculation formula is as follows: , in, Representing nodes in a similarity graph and nodes The edge weights are used to reflect the feature vectors. and The cosine similarity is used to quantify the semantic similarity between feature vectors; Indicates the first The feature vector generated after feature extraction processing of data from a data source; Indicates the first The feature vector generated after feature extraction processing of data from a data source; Represents the inner product of eigenvectors; Represents the product of modulo lengths; Furthermore, existing graph neural networks are used to perform multi-layer iterative processing on the similarity graph, updating the enhanced feature representation of each node. Specifically, this involves the following steps: In each iteration, for each node, feature vectors from neighboring nodes are collected and weighted by edge weights to obtain the aggregated feature vector. The aggregation process introduces node degree, i.e., the sum of edge weights connected to the node. By normalizing the edge weights (i.e., dividing the edge weight by the square root of the node degree), the influence of different nodes is balanced. The aggregated feature vector undergoes a linear transformation using a weight matrix to further extract semantic information. An activation function, such as ReLU, is applied to introduce non-linearity, enhancing the ability to express complex semantic relationships. The graph neural network outputs the enhanced feature representation of each node, which integrates the semantic information of neighboring nodes, reflecting the global and local associations of the feature vector in the semantic space. The calculation formula is as follows: , in, Representing the eigenvector In the graph neural network The enhanced feature representation of the layer is initially set to ; This refers to the layer index of a graph neural network; This indicates an activation function, such as ReLU, which retains positive values and sets negative values to zero, introducing non-linearity and enhancing the expressive power of graph neural networks. Indicates a node neighboring nodes Summation is performed to aggregate the feature information of neighboring nodes; Representation and eigenvectors The set of connected neighboring nodes; This represents the normalized edge weight, used to balance the influence of different nodes; Represents a node degree of nodes, ; Represents a node The degree of the node; Representing the eigenvector In the graph neural network Enhanced feature representation of layers; This represents the aggregated feature vector; The graph neural network represents the first... The layer's weight matrix, used for linear transformation and updating the enhanced feature representation of nodes, is obtained through training and optimization of a graph neural network. The value range of each element in the weight matrix is... .
[0023] Iteration After the layer, based on nodes The final enhanced feature representation The k-means clustering algorithm is applied to output semantic clusters. ,in, Indicates the first There are 1 semantic cluster; m represents the number of semantic clusters. This refers to the total number of layers in a graph neural network.
[0024] By using similarity graphs and graph neural networks, we can capture deep semantic relationships between feature vectors, enhance the discriminative power of feature representations, provide clear semantic grouping for visualization, automatically generate semantic clusters from high-dimensional features without manual intervention, and reduce the complexity of multi-business data analysis. S2. Based on semantic clusters and data foundation, cluster enhancement, dimensionality reduction and graphic rendering are performed through multi-business data visualization generation algorithms to generate visualized graphics of multi-business data. Based on semantic clusters and data foundation, a multi-business data visualization generation algorithm is used to dynamically perform cluster enhancement, dimensionality reduction and graphics rendering to generate visualized graphics of multi-business data. For each semantic cluster, intra-cluster cohesion is calculated based on the cosine similarity between feature vectors in the data foundation. This reflects the semantic closeness of feature vectors in high-dimensional space. The intra-cluster cohesion is calculated by averaging the cosine similarities between all feature vectors within the cluster. Specifically, it iterates through each pair of feature vectors within the cluster, calculates the cosine similarity, sums them, and divides by the total number of feature vector pairs within the cluster. This factorial form of the number of feature vectors within the cluster reflects the semantic tightness of the feature vectors within the cluster. The calculation formula is as follows: , in, Indicates the first Intra-cluster cohesion is used to quantify the semantic tightness of feature vectors within a cluster. Indicates the first The number of eigenvectors of a cluster is used to normalize the calculation of cluster cohesion; This represents the sum of the cosine similarities of feature vector pairs within a cluster.
[0025] To enhance inter-cluster discriminability, the minimum similarity between each pair of different clusters is calculated. For each pair of clusters, all feature vectors of one cluster and all feature vectors of the other cluster are traversed, and the cosine similarity is calculated. The minimum value among all cosine similarities is taken as the inter-cluster separation degree, reflecting the weakest degree of semantic association between the two clusters. The calculation formula is as follows: , in, Indicates the first Cluster and the first Inter-cluster separation is used to quantify inter-cluster differences; This represents the minimum value operation between clusters.
[0026] Based on intra-cluster cohesion and inter-cluster separation, the similarity weights between feature vectors are dynamically adjusted. For feature vector pairs within the same cluster, the cosine similarity is multiplied by an enhancement factor, which consists of intra-cluster cohesion and an enhancement adjustment parameter, to amplify the similarity of feature vectors within the cluster and strengthen intra-cluster compactness. For feature vector pairs from different clusters, the cosine similarity is multiplied by a reduction factor, which consists of inter-cluster separation and a reduction adjustment parameter, to reduce the similarity of feature vectors between clusters and highlight inter-cluster differences. The adjusted similarity weights retain the original similarity topology while enhancing intra-cluster aggregation and inter-cluster separation effects through dynamic weighting. The calculation formula is as follows: , in, Represents nodes in the adjusted similarity graph and nodes The edge weights, i.e., similarity weights; This represents the enhancement adjustment parameter, used to control the intensity of intra-cluster similarity enhancement. It is determined based on expert experience and its value range is [range missing]. ; Indicates the enhancing factor; This represents the reduction adjustment parameter, used to control the intensity of inter-cluster similarity reduction. It is determined based on expert experience and its value range is [range missing]. ; Indicates the weakening factor; Indicates the first A semantic cluster.
[0027] The Uniform Manifold Approximation and Projection (UMAP) algorithm is used to map high-dimensional feature vectors to a low-dimensional space (two-dimensional or three-dimensional) to generate visual coordinates.
[0028] In the low-dimensional space, initial low-dimensional coordinates (2D or 3D) are randomly generated for each feature vector as the starting point for UMAP algorithm optimization. The low-dimensional coordinates are optimized by minimizing the difference between the high-dimensional similarity and low-dimensional similarity distributions. The optimization objective is to minimize the cross-entropy loss between high-dimensional and low-dimensional similarity. The low-dimensional coordinates are iteratively updated using the existing stochastic gradient descent method. The high-dimensional similarity comes from the adjusted similarity weights, and the low-dimensional similarity is based on the Euclidean distance of the low-dimensional coordinates. Two hyperparameters are introduced to control the distance scaling and shape, respectively. After multiple iterations, the low-dimensional coordinates are obtained, and each low-dimensional coordinate corresponds to a high-dimensional feature vector, preserving the enhanced intra-cluster and inter-cluster semantic relationships.
[0029] The optimization process is expressed by the following formula: , in, This indicates a minimize operation; This represents the first term of the cross-entropy, used to measure the difference in similarity between high-dimensional and low-dimensional dimensions; This represents the second term of the cross-entropy, used to ensure the separation of dissimilar points; Represent the natural logarithm function; Representing nodes in a similarity graph and nodes The low-dimensional similarity weight is calculated using the following formula: , in, and This represents the hyperparameters of the UMAP algorithm, used to control the low-dimensional similarity distribution, with values ranging from [value range missing]. ; Represents the Euclidean distance of low-dimensional coordinates; Indicates the first Low-dimensional coordinates of data from a single data source; Indicates the first Low-dimensional coordinates of data from a data source.
[0030] Depending on the application scenario, such as finance, healthcare, or industry, a suitable visualization format can be selected from a predefined template library. This includes scatter plots to display cluster distribution; relationship graphs to present inter- and intra-cluster connections; and heatmaps to reflect similarity strength. Template parameters define the graph type and attributes, such as color, shape, and edge thickness, and are automatically matched based on user needs or data characteristics (such as cluster size and semantic complexity) to ensure adaptability to multiple domain scenarios. Using low-dimensional coordinates, semantic clusters, and template parameters, mature graphics libraries such as D3.js, ECharts, and Plotly are used to generate visualizations.
[0031] In summary, a method for generating multi-business data visualizations using AI agents has been developed.
[0032] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0033] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0034] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method for generating multi-business data visualization using AI intelligent agents, characterized in that, Includes the following steps: S1. Obtain data from various data sources and extract features to obtain feature vectors. Integrate all feature vectors into a data base and extract semantic relationships from the data base using a semantic association graph clustering algorithm to generate semantic clusters. S2. Based on semantic clusters and data foundation, cluster enhancement, dimensionality reduction, and graphic rendering are performed through multi-business data visualization generation algorithms to generate visualized graphics of multi-business data.
2. The method for generating multi-business data visualization using AI intelligent agents according to claim 1, characterized in that, S1 specifically includes: In the implementation of the semantic association graph clustering algorithm, cosine similarity is calculated for each pair of feature vectors in the data base; cosine similarity is calculated by dividing the inner product of feature vectors by the product of their moduli, generating edge weights, i.e., similarity weights.
3. The method for generating multi-business data visualization using AI intelligent agents according to claim 2, characterized in that, S1 specifically includes: A similarity graph is constructed based on similarity weights. Nodes in the similarity graph represent feature vectors, and edges connect each pair of nodes. The edge weight is the corresponding cosine similarity.
4. The method for generating multi-business data visualization using AI intelligent agents according to claim 3, characterized in that, S1 specifically includes: A graph neural network is used to perform multi-layer iterative processing on the similarity graph, and the final enhanced feature representation of the nodes is output. Based on the final enhanced feature representation of the nodes, a semantic cluster is output.
5. The method for generating multi-business data visualization using AI intelligent agents according to claim 1, characterized in that, S2 specifically includes: Based on the cosine similarity between feature vectors in the data base, the cosine similarity between all feature vectors within each semantic cluster is averaged to obtain the intra-cluster cohesion.
6. The method for generating multi-business data visualization using AI intelligent agents according to claim 1, characterized in that, S2 specifically includes: For each pair of semantic clusters, iterate through all feature vectors of one cluster and all feature vectors of the other cluster, calculate the cosine similarity, and take the minimum value among all cosine similarities as the inter-cluster separation degree.
7. The method for generating multi-business data visualization using AI intelligent agents according to claim 6, characterized in that, S2 specifically includes: Based on intra-cluster cohesion and inter-cluster separation, the similarity weights between feature vectors are dynamically adjusted; for feature vector pairs within the same cluster, the cosine similarity is multiplied by an enhancement factor; for feature vector pairs from different clusters, the cosine similarity is multiplied by a weakening factor to obtain the adjusted similarity weights.
8. The method for generating multi-business data visualization using AI intelligent agents according to claim 7, characterized in that, S2 specifically includes: Based on the adjusted similarity weights, a uniform manifold approximation and projection algorithm is used to map high-dimensional feature vectors to a low-dimensional space to generate visual coordinates. In the low-dimensional space, initial low-dimensional coordinates are randomly generated for each feature vector as the starting point for the optimization of the uniform manifold approximation and projection algorithm. The low-dimensional coordinates are optimized by minimizing the difference between the high-dimensional similarity and the low-dimensional similarity distribution.