A non-parametric bayesian based relational graph data clustering method and system

Abstract

The application relates to the technical field of relationship graph data clustering, and provides a relationship graph data clustering method and system based on a non-parametric Bayesian method, which comprises the following steps: obtaining relationship graph data; initializing model parameters according to a probabilistic graph model; using multiple sampling algorithms to iteratively sample samples of the model parameters based on the relationship graph data; selecting a cluster division sample according to a maximum likelihood function mode based on the model parameters obtained through sampling, so as to obtain a clustering result; and the prior probability of the cluster division sample is a non-parametric Bayesian prior. The number of clusters can be automatically inferred in the parameter inference process, and manual setting is not needed.

Description

Technical Field

[0001] This invention belongs to the field of relational graph data clustering technology, and particularly relates to a relational graph data clustering method and system based on nonparametric Bayes. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As the technological barriers to big data continue to decrease, various industries are undergoing digital transformation. Data plays a central role in this transformation, and many enterprises and organizations possess vast amounts of data. However, to extract information from these massive datasets, leading internet companies are choosing to build various knowledge graphs, using graph databases to store various entities and relationships, and leveraging machine learning and deep learning methods to mine the relational graph data. This data mining approach, combining knowledge graphs with machine learning, is not only a key point for enterprises, organizations, and industries to integrate with digitalization and intelligentization, but it can also generate valuable business intelligence.

[0004] There are many algorithms for mining graph data, among which graph node clustering algorithms can divide nodes in a graph into subsets with similar characteristics and functions, making them a crucial machine learning algorithm for graph data. By clustering nodes in a relationship graph, hidden structures and features can be uncovered, which are often impossible to obtain through manual observation and statistical data. For example, clustering a person-event relationship graph can reveal the intersection of people's activities; clustering a user-item relationship graph can simultaneously discover different user preferences and different item types; and clustering an entity-attribute relationship graph can reveal the strength of relationships between entities.

[0005] Nodes in a graph can represent things of the same type (homogeneous nodes), such as nodes representing people in a co-occurrence relationship; or they can represent things of different types (heterogeneous nodes), such as nodes representing users and items in a user-item purchase relationship, where nodes of the same type do not have a purchase relationship. Common graph node clustering methods include spectral clustering, random walk-based methods, and methods based on unsupervised deep learning + k-means (or + spectral clustering), but these methods are not specifically designed for graph data and have limitations when clustering graph data.

[0006] First, the above model requires setting the number of clusters in advance during the clustering process of relational graph data. It lacks the ability to automatically discover the number of clusters. The estimated number of clusters needs to be given before the clustering node. This simplifies the clustering process of the model, but to find a reasonable number of clusters, it is necessary to use methods such as minimizing the description length and Bayesian model selection, which not only increases the implementation complexity, but also results in poor clustering effect.

[0007] Secondly, while the above models perform well in clustering homogeneous nodes, they lack consideration for the relationships between heterogeneous nodes, such as whether they participate in (an event), whether they purchase (an item), or whether they possess (a certain characteristic). This results in poor clustering performance on such datasets. Summary of the Invention

[0008] To address the technical problems mentioned above, this invention provides a nonparametric Bayesian-based relational graph data clustering method and system. This method utilizes a nonparametric Bayesian approach to generate node cluster partitions, and can automatically infer the number of clusters during parameter inference without requiring manual settings.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] The first aspect of this invention provides a nonparametric Bayesian-based graph data clustering method, comprising:

[0011] Obtain relationship graph data;

[0012] Initialize the model parameters based on the probabilistic graphical model;

[0013] Based on the relational graph data, various sampling algorithms are used to iteratively sample the model parameters;

[0014] Based on the model parameters obtained from sampling, clusters are selected to divide the samples according to the method of maximizing the likelihood function, and the clustering results are obtained.

[0015] The prior probability of cluster partitioning samples is a nonparametric Bayesian prior.

[0016] Furthermore, the relational graph data is preprocessed before iteratively sampling the model parameters;

[0017] The preprocessing includes removing orphaned nodes, redundant self-loops, and converting relational data into a uniform format.

[0018] Furthermore, the process of sampling the parameters of the iterative sampling model is as follows:

[0019] Based on hyperparameters, the Metropolis-Hasting sampling algorithm or the Gibbs sampling algorithm is used to sample clusters to divide samples.

[0020] Based on cluster-based sample partitioning, hyperparameters are sampled using auxiliary variable methods or slice sampling methods.

[0021] Based on the standardized mutual information and the number of iterations, determine whether to exit the sampling iteration.

[0022] Furthermore, the relationship graph data is either homogeneous node relationship graph data or heterogeneous node relationship graph data.

[0023] Furthermore, based on the likelihood probability, the cluster partition sample that maximizes the log-likelihood is selected as the clustering result.

[0024] Furthermore, by utilizing the conjugate relationship between the multinomial distribution and the Dirichlet distribution, the connection probability vector in the probabilistic graphical model is eliminated through integration.

[0025] A second aspect of the present invention provides a nonparametric Bayesian-based relational graph data clustering system, comprising:

[0026] The data acquisition module is configured to acquire relationship graph data.

[0027] The initialization module is configured to initialize model parameters based on the probabilistic graphical model.

[0028] The iterative sampling module is configured to iteratively sample the model parameters based on the relation graph data and using multiple sampling algorithms.

[0029] The clustering module is configured to select clusters based on the sampled model parameters and maximize the likelihood function to obtain the clustering results.

[0030] The prior probability of cluster partitioning samples is a nonparametric Bayesian prior.

[0031] Furthermore, it also includes a preprocessing module configured to preprocess the relational graph data before iteratively sampling the model parameters;

[0032] The preprocessing includes removing orphaned nodes, redundant self-loops, and converting relational data into a uniform format.

[0033] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the nonparametric Bayesian-based relational graph data clustering method described above.

[0034] A fourth aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a nonparametric Bayesian-based relational graph data clustering method as described above.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] This invention provides a nonparametric Bayesian-based clustering method for relational graph data. It uses a nonparametric Bayesian method to generate cluster partitions of nodes, and can automatically infer the number of clusters during the parameter inference process without the need for manual setting.

[0037] This invention provides a nonparametric Bayesian clustering method for relational graph data, which uses probability vectors to describe the connectivity characteristics of nodes within a cluster, enabling the model to be used for clustering relational graph data with both homogeneous and heterogeneous nodes. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 This is a flowchart of a nonparametric Bayesian-based relational graph data clustering method according to Embodiment 1 of the present invention;

[0040] Figure 2 This is a schematic diagram of the probability graph model of Embodiment 1 of the present invention;

[0041] Figure 3 This is a flowchart of the relationship graph data clustering process according to Embodiment 1 of the present invention. Detailed Implementation

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, 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.

[0044] Example 1

[0045] This embodiment provides a nonparametric Bayesian-based clustering method for relational graph data, such as... Figure 1 As shown, it includes the following steps:

[0046] Step 1: Establish a probabilistic graphical model.

[0047] like Figure 2 As shown, this invention proposes a probabilistic graphical model to describe the generation process of graph data. This model utilizes a non-parametric Bayesian method to generate node clusters, automatically inferring the number of clusters during parameter inference without requiring manual settings. Simultaneously, the model uses probability vectors to describe the connectivity characteristics of nodes within clusters, enabling it to be used for clustering graph data with both homogeneous and heterogeneous nodes.

[0048] The model includes the observable variable r.(n) And latent variables (i.e. model parameters) p, θ, c, a and optional parameter b. Figure 2 The quantities in the circles with a medium gray background represent observable random quantities, while the quantities in the circles with a white background represent model parameters.

[0049] Where, r (n A relation (n) represents relational graph data with n instances or objects, such as co-occurrence, participation, purchase, or ownership relationships. Each relation instance has two entities as nodes, and the edges connecting the two nodes indicate that the two entities conform to the relation. Vector c represents the clustering of all nodes. Assuming there are currently K clusters with m nodes, then c... i =k, where i is a positive integer less than or equal to m, and k is a positive integer less than or equal to K. The prior probability of c is a nonparametric Bayesian prior, characterized by c being generated in a constructive manner. During the construction process, the maximum value of the elements in c represents the number of clusters K, which can vary with r. (n) The number of nodes should be increased appropriately. The hyperparameters of the nonparametric Bayesian prior can be set to 'a', or 'a' and 'b', based on the statistical characteristics of the sample.

[0050] Vector p represents a shared connection probability vector within a cluster. With K clusters, there are K different probability vectors. Nodes within the same cluster will establish relationships with other nodes with similar probabilities. For example, endpoint security products are likely to have sub-functions such as process monitoring, registry monitoring, critical file monitoring, and storage device monitoring; network security products are likely to have sub-functions such as traffic monitoring, log auditing, and network port monitoring. Therefore, endpoint security products and network security products will have different functions with varying probabilities. However, similar products designed by different manufacturers will have different sub-functions, and endpoint security and network security products also have a small probability of having similar sub-functions. Vector p is a probability vector whose prior is a multivariate symmetric Dirichlet distribution with parameter θ.

[0051] Step 2: Data cleaning and preprocessing, initialization of model parameters, sampling algorithm iteration, determination of reasonable results and output.

[0052] like Figure 3 As shown, this invention proposes a Markov Chain Monte Carlo (MCMC) sampling algorithm based on the above probabilistic graphical model. This algorithm uses various sampling algorithms, including Gibbs sampling, auxiliary variable sampling, and slice sampling, to sample the model parameters. Then, it selects the most suitable cluster partitioning sample based on maximizing the likelihood function. Specifically, it includes the following steps:

[0053] (1) Obtain relation graph data r (n) The graph data is preprocessed, including removing isolated nodes and redundant self-loops, and converting the original relational data into a unified format.

[0054] Relationship graph data can be homogeneous node relationship graph data. For example, if the relationship graph data is dolphin social relationship graph data, the nodes are dolphins, and the edge connecting two nodes represents the social relationship between the two dolphins. If the relationship graph data is people co-occurrence relationship graph data, the nodes are people, and the edge connecting two nodes represents the co-occurrence relationship between the two people.

[0055] Relationship graph data can be heterogeneous node relationship graph data. For example, if the relationship graph data is about women and social event participation, some nodes are women and some nodes are social events, and the edge between two nodes represents the participation relationship between women and social events. If the relationship graph data is about user-item purchase relationship graph data, the nodes are divided into two categories: users and items. There are purchase relationships between nodes of different categories, but no purchase relationships between nodes of the same category.

[0056] (2) Based on the probability graphical model ( Figure 2 Initialize the model parameters, including θ, c, a (and b) and the number of sampling iterations J. The initialization process for the cluster partitioning vector c is as follows:

[0057] (201) Given m nodes, the initial cluster size is K = log(m),

[0058] (202) Randomly assign m nodes to K clusters to obtain a cluster partition vector. The i-th dimension of the vector represents the cluster number (i.e., an element) to which the i-th node belongs (this number can be a number, symbol, or string, such as c). 0 =[3,2,1...]、c 0 =["c","b","a"...]、c 0 =["customer","dairy","meat"...]), this vector uses c 0 express.

[0059] (3) Based on the hyperparameters, Gibbs sampling is performed according to the conditional probability (1) to obtain the cluster partitioning vector c after the j-th iteration. j The sample of each element is used to distinguish the intermediate results of the Gibbs sampling process from the cluster partitioning result after the j-th sampling. j , use c (m) This represents the intermediate results of the Gibbs sampling process (m indicates that there are m nodes in a graph). c (m) Ignore all elements after the i-th element, such as c. (m) =[2,3,1...], Here, i = 1.

[0060]

[0061] Among them, c i c (m) Cluster partitioning of node i in the data, sampling c according to conditional probability (1) i After that, c i The value of , i.e., the cluster number, needs to be updated synchronously to c. (m) The i-th dimension, the updated c (m) Used for the next sampling. j This represents the cluster partitioning of all nodes after the j-th iteration. (m) c (m) There are m elements, and similarly r (n) This represents n relations in a relation graph.

[0062] Alternatively, the Metropolis-Hasting (MH) sampling algorithm can be used to partition sample clusters into c. j .

[0063] (4) Based on cluster partitioning, sample hyperparameters a (and b) using auxiliary variable method or slice sampling method according to conditional probability (2).

[0064] p(a|r (n) ,c (m) ,θ)=p(c (m) |a)p(a) or p(a,b|r (n) ,c (m) ,θ)=p(c (m) |a,b)p(a,b) (2)

[0065] (5) Based on the conditional probability (3), use the slice sampling method to sample the hyperparameter θ.

[0066] p(θ|r (n) ,c (m) ,a,b)=p(r (n) |c (m) ,θ)p(θ) (3)

[0067] (6) By calculating the sample size c of this cluster partitioning j Sample c from the previous cluster partition j-1 The standardized mutual information (NMI) is used to determine whether it is less than the threshold. If it is less than the threshold, the sampling iteration can be exited and the process can proceed to step (8) to avoid overfitting.

[0068] (7) The iteration number j = j + 1. Determine whether the iteration number j is less than J. If yes, return to step (3); otherwise, proceed to step (8).

[0069] (8) Based on the likelihood probability (4), divide the samples c into all clusters. jIn the selection, choose c that maximizes the log-likelihood. j Marked as c o , where o is a positive integer less than or equal to J.

[0070] p(r (n) ,c (m) |a,b,θ)=p(r (n) |c (m) ,θ)p(c (m) |a,b) (4)

[0071] (9) Output or save clustering results c o .

[0072] If the relationship graph data consists of homogeneous nodes, the clustering result will divide the nodes into several clusters. For example, if the relationship graph data is dolphin social relationship graph data, each cluster in the clustering result will contain several dolphins (nodes). If the relationship graph data is co-occurrence relationship graph data of people, each cluster in the clustering result will contain several people.

[0073] If the relationship graph data consists of homogeneous nodes, the clustering result will divide each type of node into several clusters. For example, if the relationship graph data consists of women and social event participation, each cluster in the clustering result will contain several women or several social events. If the relationship graph data consists of user-item purchase relationship graph data, each cluster in the clustering result will contain several users or several items.

[0074] Because in generating r (n) In the process, vector p is the parameter of a multinomial distribution. Therefore, the conjugate relationship between Dirichlet and multinomial distributions can be used to eliminate the parameter p by integrating over it. In the actual MCMC sampling process, p does not need to be considered.

[0075] Two datasets were used in the experiment: one consisted of homogeneous node relationship graphs, including the "dolphin social relationship" graph; the other consisted of heterogeneous node graphs, including the relationship graph between women and social event participation. Comparison methods included classic spectral clustering, Infomap based on random walks, and a deep latent feature relationship model based on autoencoders. This model used MCMC with 500 iterations, and converged after approximately 50 iterations on the four datasets.

[0076] After clustering homogeneous node relationship graph data, cluster labels for all nodes can be obtained. These label sequences are then compared with the actual label sequences, and the difference between the two is calculated using Normalized Mutual Information (NMI). If the NMI value is 100%, the clustering result perfectly matches the actual result; the smaller the NMI value, the lower the similarity between the clustering result and the actual result. Experiments show that spectral clustering algorithms, Infomap, and deep latent feature relationship models, after being informed of the actual number of clusters K, find cluster partitions with NMI values ​​ranging from 21% to 67% relative to the actual value. In contrast, the model of this invention, with the actual number of clusters K, achieves an NMI value as high as 97% in dolphin social relationship data, significantly exceeding the comparative models. Therefore, this invention can be used for clustering homogeneous node relationship graphs and can automatically detect the number of clusters.

[0077] After clustering heterogeneous node relationship graph data, cluster labels for all nodes can be obtained, and each cluster label has at least two elements. Nodes with the same label have no relational connections. Taking the "ladies and social events" graph data as an example, after clustering, four cluster labels can be obtained, representing two types of ladies and two types of events. Ladies of one type tend to participate more in one type of event and less in the other. The experiment uses the error rate (misclassification rate) of classifying a certain type of node into another type of cluster as a metric for evaluating the model. The experiment shows that spectral clustering algorithms, Infomap, and deep latent feature relationship models, even when the true number of clusters K is known to the model, still have an error rate (misclassification rate) of 12%-33% when classifying a certain type of node into another type of cluster. In contrast, the model of this invention, without knowing the true number of clusters K, has an error rate of only 1% in the ladies and social event participation relationship data. Therefore, this invention can be used for clustering heterogeneous node relationship graphs and can automatically discover the number of clusters.

[0078] Example 2

[0079] This embodiment provides a nonparametric Bayesian-based graph data clustering system, which specifically includes the following modules:

[0080] The data acquisition module is configured to acquire relationship graph data.

[0081] The initialization module is configured to initialize model parameters based on the probabilistic graphical model.

[0082] The iterative sampling module is configured to iteratively sample the model parameters based on the relation graph data and using multiple sampling algorithms.

[0083] The clustering module is configured to select clusters based on the sampled model parameters and maximize the likelihood function to obtain the clustering results.

[0084] The prior probability of cluster partitioning samples is a nonparametric Bayesian prior.

[0085] The preprocessing module is configured to preprocess the relational graph data before iteratively sampling the model parameters.

[0086] The preprocessing includes removing orphaned nodes, redundant self-loops, and converting relational data into a uniform format.

[0087] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation processes are the same, so they will not be repeated here.

[0088] Example 3

[0089] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in a nonparametric Bayesian-based relational graph data clustering method as described in Embodiment 1 above.

[0090] Example 4

[0091] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the nonparametric Bayesian-based relational graph data clustering method described in Embodiment 1 above.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0097] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A nonparametric Bayesian-based clustering method for relational graph data, characterized in that, include: Obtain relationship graph data; The relationship graph data is either homogeneous node relationship graph data or heterogeneous node relationship graph data; Initialize the model parameters based on the probabilistic graphical model; Based on the relational graph data, various sampling algorithms are used to iteratively sample the model parameters; The various sampling algorithms include Gibbs sampling and Metropolis-Hasting sampling. The samples are divided into clusters using the aforementioned multiple sampling algorithms; Based on cluster-based sample partitioning, hyperparameters are sampled using auxiliary variable method or slice sampling method according to conditional probability. By calculating the standardized mutual information between the current cluster partitioning sample and the previous cluster partitioning sample, it is determined whether the information is less than a threshold. If the information is less than the threshold, the sampling iteration can be terminated. Based on the model parameters obtained from sampling, clusters are selected to divide the samples according to the method of maximizing the likelihood function, and the clustering results are obtained. Among them, the prior probability of cluster partitioning samples is a nonparametric Bayesian prior; For clustering of heterogeneous node relationship graph data, the relationship graph data is user-item purchase relationship graph data. The nodes are divided into two categories: users and items. There are purchase relationships between nodes of different categories, but there are no purchase relationships between nodes of the same category.

2. The nonparametric Bayesian-based relational graph data clustering method as described in claim 1, characterized in that, Preprocess the graph data before iteratively sampling the model parameters; The preprocessing includes removing orphaned nodes, redundant self-loops, and converting relational data into a uniform format.

3. The nonparametric Bayesian-based relational graph data clustering method as described in claim 1, characterized in that, The process of iteratively sampling model parameters is as follows: Based on hyperparameters, the Metropolis-Hasting sampling algorithm or the Gibbs sampling algorithm is used to sample clusters to divide samples. Based on cluster-based sample partitioning, hyperparameters are sampled using auxiliary variable methods or slice sampling methods. Based on the standardized mutual information and the number of iterations, determine whether to exit the sampling iteration.

4. The nonparametric Bayesian-based relational graph data clustering method as described in claim 1, characterized in that, Based on the likelihood probability, the cluster partition samples that maximize the log-likelihood are selected as the clustering results.

5. The nonparametric Bayesian-based relational graph data clustering method as described in claim 1, characterized in that, By utilizing the conjugate relationship between the multinomial distribution and the Dirichlet distribution, the connection probability vector in the probabilistic graphical model is eliminated through integration.

6. A relational graph data clustering system based on nonparametric Bayesian methods, characterized in that, include: The data acquisition module is configured to acquire relationship graph data. The relationship graph data is either homogeneous node relationship graph data or heterogeneous node relationship graph data; The initialization module is configured to initialize model parameters based on the probabilistic graphical model. The iterative sampling module is configured to iteratively sample the model parameters based on the relation graph data and using multiple sampling algorithms. The various sampling algorithms include Gibbs sampling and Metropolis-Hasting sampling. The samples are divided into clusters using the aforementioned multiple sampling algorithms; Based on cluster-based sample partitioning, hyperparameters are sampled using auxiliary variable method or slice sampling method according to conditional probability. By calculating the standardized mutual information between the current cluster partitioning sample and the previous cluster partitioning sample, it is determined whether the information is less than a threshold. If the information is less than the threshold, the sampling iteration can be terminated. The clustering module is configured to select clusters based on the sampled model parameters and maximize the likelihood function to obtain the clustering results. Among them, the prior probability of cluster partitioning samples is a nonparametric Bayesian prior; For clustering of heterogeneous node relationship graph data, the relationship graph data is user-item purchase relationship graph data. The nodes are divided into two categories: users and items. There are purchase relationships between nodes of different categories, but there are no purchase relationships between nodes of the same category.

7. The relationship graph data clustering system based on nonparametric Bayes as described in claim 6, characterized in that, It also includes a preprocessing module, which is configured to preprocess the relational graph data before iteratively sampling the model parameters; The preprocessing includes removing orphaned nodes, redundant self-loops, and converting relational data into a uniform format.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the nonparametric Bayesian relational graph data clustering method as described in any one of claims 1-5.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the nonparametric Bayesian relational graph data clustering method as described in any one of claims 1-5.

Images