Overlapped clustering method and system based on bipartite graph batch Chinese restaurant process and medium

Abstract

The invention provides an overlapping clustering method and system based on a bipartite graph batch Chinese restaurant process and a medium, and relates to the technical field of machine learning, and the method comprises the steps: receiving incremental data of a current batch, and generating a candidate new cluster according to the compatibility, novelty and batch drift degree of a sample and a historical cluster in the incremental data; respectively calculating scores of the samples distributed to the existing clusters and the new clusters by using a batch Chinese restaurant process; constructing a dynamic bigraph and an edge weight of the nodes and cluster nodes; performing cutting on the local sub-graph influenced by the current batch; determining a node type based on the edge weight, the bridging entropy, the significant connection number and the attribution weight; determining a cluster state according to the stability, activeness and dispersion of the cluster; and fusing multi-source evidences, fusing the multi-source evidences with historical membership degrees after bidirectional propagation, calculating an adaptive threshold according to node types, bridging entropy and cluster states, and outputting an overlapped cluster set of each sample. According to the method, the problem that dynamic nature, overlapping property and high efficiency are difficult to consider in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to an overlapping clustering method, system, and medium based on a bipartite graph-based batch processing of Chinese restaurants. Background Technology

[0002] Clustering analysis is a fundamental unsupervised learning method in data mining, pattern recognition, and complex network analysis. It primarily groups similar samples into the same cluster based on their similarity or correlation. Existing clustering methods, such as K-means, hierarchical clustering, and spectral clustering, typically employ mutually exclusive partitioning, meaning each sample belongs to only one cluster. These methods are well-suited for scenarios with relatively stable data structures and clear category boundaries. However, in practical applications, samples often possess multiple attributes or semantic features and may belong to multiple clusters simultaneously. Therefore, traditional mutually exclusive clustering methods struggle to accurately reflect the true structure of complex data objects.

[0003] To address the aforementioned issues, existing research has proposed overlapping clustering methods, allowing the same sample to belong to multiple clusters simultaneously, thus better reflecting the cross-associations between samples. Existing overlapping clustering methods mainly include matrix factorization-based methods, fuzzy membership-based methods, graph structure-based methods, and probabilistic generative model-based methods. These methods have certain application value in fields such as social network analysis, text topic mining, bioinformatics, and recommender systems. However, most existing overlapping clustering methods are primarily designed for static datasets and typically require offline clustering calculations under full data conditions, making them difficult to adapt to dynamic scenarios where data continuously arrives and structures constantly change.

[0004] With the development of data stream processing and online learning technologies, dynamic clustering methods for incremental or streaming data have emerged in existing technologies. These methods can update existing clustering results as new samples arrive, reducing the computational overhead of repeated training to some extent and improving adaptability to data evolution. However, most existing dynamic clustering methods still primarily focus on non-overlapping clustering, typically assigning newly arrived samples to a single cluster or only supporting structural changes such as cluster generation, extinction, splitting, and merging. They struggle to further characterize the overlapping relationships where samples simultaneously belong to multiple clusters. Therefore, existing dynamic clustering methods still have shortcomings when handling data objects that simultaneously possess dynamic and overlapping characteristics.

[0005] Graph-based clustering methods have been widely used in recent years due to their ability to effectively represent relationships between samples. Among them, bipartite graphs can effectively represent the associations between two types of nodes, making them suitable for modeling relationships between samples and clusters, samples and features, or samples and representative nodes. Therefore, they have strong expressive power in cluster analysis and community detection. Existing bipartite graph clustering methods typically achieve sample category discovery by partitioning the bipartite graph, while some methods utilize the connectivity relationships within the bipartite graph for overlapping community identification. These methods have certain advantages in preserving the original relational structure and reducing information loss. However, most existing bipartite graph clustering methods focus on static graph structures, usually requiring the pre-construction of a complete graph model before obtaining clustering results through graph partitioning or graph propagation. This makes it difficult to adapt to the dynamic updating requirements of the graph structure under conditions of continuously arriving incremental samples.

[0006] The Chinese restaurant process and its extended model in nonparametric Bayesian methods have the advantage of adaptive cluster number generation, avoiding the problem of pre-setting the number of clusters required in traditional clustering methods. Existing research has used extended models of the Chinese restaurant process for dynamic clustering and dependency modeling to achieve adaptive adjustment of cluster structure as data changes. However, most existing methods focus on the probabilistic inference process of sample partitioning, mainly outputting the cluster affiliation of samples. They lack comprehensive technical solutions for how to combine bipartite graph structures to express the dynamic relationships between samples and clusters, how to optimize cluster structure through graph cutting, and how to achieve overlapping clustering construction through key node selection, label propagation, and label fusion. Furthermore, some studies have high computational complexity during the inference process, making it difficult to meet the real-time and scalability requirements of large-scale incremental data scenarios.

[0007] In summary, existing technologies still have shortcomings in dynamic overlapping clustering: First, most existing overlapping clustering methods are based on static batch processing models, making it difficult to adapt to the dynamic evolution of cluster structures under incremental data conditions; second, although existing dynamic clustering methods can handle data stream updates, they often struggle to effectively characterize the overlapping membership relationships of multiple clusters; third, while existing bipartite graph clustering methods can utilize graph structure information for sample partitioning, they generally lack the ability to combine with dynamic incremental learning mechanisms, making it difficult to achieve unified processing of dynamic bipartite graph construction, graph cutting, key node selection, and subsequent overlapping cluster generation; fourth, some methods still rely on manually preset cluster numbers, empirical thresholds, or computationally expensive inference mechanisms, making it difficult to meet the application needs of complex dynamic scenarios. Therefore, it is necessary to propose a new clustering construction method to address the difficulty in simultaneously achieving dynamism, overlap, and efficiency in existing technologies. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide an overlapping clustering method, system and medium based on bipartite graph batch Chinese restaurant process, so as to solve the problem that dynamics, overlap and efficiency are difficult to balance in the prior art.

[0009] In a first aspect, embodiments of the present invention provide an overlapping clustering method based on a bipartite graph-based batch Chinese restaurant process, comprising: Receive incremental data from the current batch, and generate candidate new clusters based on the compatibility, novelty, and batch drift of the samples in the incremental data with historical clusters; The batch Chinese restaurant process is used to calculate the scores of the samples assigned to existing clusters and new clusters respectively, and overlapping candidate clusters are retained based on the scores; Construct a dynamic bipartite graph of nodes and cluster nodes and edge weights, calculate bridging entropy based on edge weights to identify bridging edges; perform cutting on local subgraphs affected by the current batch, and apply reversible retention scores to cut bridging edges; Node types are determined based on edge weights, bridging entropy, significant connectivity, and affiliation weights; cluster states are determined based on cluster stability, activity, and dispersion. Multi-source evidence is fused, and the evidence propagation intensity is adjusted according to node type and cluster state. After bidirectional propagation, it is fused with historical membership degree, and an adaptive threshold is calculated based on node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

[0010] Optionally, the process of generating candidate new clusters includes: Determine whether a new pattern not seen in the history appears in the current batch of samples. If so, determine the number of new clusters based on the compatibility of the samples with historical clusters, sample novelty, overall batch drift, and risk of historical cluster splitting. Initialize the new cluster centers from samples that meet the novelty threshold and local support.

[0011] Optionally, the batch Chinese restaurant process is used to calculate the scores of sample assignment to existing and new clusters, and overlapping candidate clusters are retained based on the scores, including: Intra-batch similarity is calculated based on feature similarity and neighborhood structure similarity between samples; The first score for assigning a sample to each existing cluster and the second score for assigning a sample to each candidate new cluster are calculated. The score for assigning a sample to an existing cluster incorporates the historical size of the cluster, the compatibility between the sample and the cluster, intra-batch collaboration, cross-batch memory, and cluster state mapping adjustment. The score for assigning a sample to a new cluster incorporates the adaptive concentration parameter, the sample novelty, and the local support of the sample to the candidate new cluster. Normalize the first and second scores to obtain the probability that the sample belongs to each cluster; Based on the relative probability threshold, overlapping candidate clusters are selected and retained to determine the multi-cluster overlapping membership relationship of the samples.

[0012] Optionally, construct a dynamic bipartite graph and edge weights for nodes and cluster nodes, including: Edge weights are determined by a combination of the current node's affiliation strength, sample-cluster compatibility, historical connection memory, and bridging preservation term. Edge states are classified according to their weight distribution, including core edges, bridging edges, and drifting edges.

[0013] Optionally, a slicing process may be performed on local subgraphs affected by the current batch, including: Based on the bridging entropy of the samples, identify the samples and bridging edges located in the overlapping boundary region; When a bridging edge is cut, an additional cost is imposed on the bridging edge based on the historical edge weight, cluster stability, and current edge weight; if the cut bridging edge satisfies the reversibility condition in subsequent consecutive batches, the connection of the bridging edge is restored.

[0014] Optionally, the reversible retention score is obtained by weighted fusion of historical edge weights, cluster stability, and current edge weights.

[0015] Optionally, the process of determining the cluster state includes: Clusters whose activity increases in consecutive batches and whose cluster center changes are below a threshold are kept as active clusters; clusters whose sample dispersion increases in consecutive batches and whose cluster center changes are above a threshold are marked as clusters to be split; two clusters whose shared sample ratio is above a threshold and whose cluster center distance is below a threshold are merged; clusters whose activity is below a threshold in consecutive batches are marked as dormant clusters.

[0016] Optionally, an adaptive threshold can be calculated based on node type, bridging entropy, and cluster state, including: The threshold for core nodes is higher than that for bridging nodes and boundary nodes; the threshold for sample nodes belonging to the cluster to be split or merged is lower than that for sample nodes belonging to the active or dormant cluster; when a sample is both a core node and belongs to the cluster to be split, the lower threshold corresponding to the cluster to be split is used.

[0017] Secondly, the present invention also provides an overlapping clustering system based on a bipartite graph batch Chinese restaurant process, the system comprising: The generation unit receives incremental data from the current batch and generates candidate new clusters based on the compatibility, novelty, and batch drift of the samples in the incremental data with historical clusters. The calculation unit is used to calculate the scores of the samples assigned to existing clusters and new clusters respectively using the batch Chinese restaurant process, and retain overlapping candidate clusters based on the scores; The cutting unit is used to construct a dynamic bipartite graph of nodes and cluster nodes and edge weights, calculates bridging entropy based on edge weights to identify bridging edges, performs cutting on local subgraphs affected by the current batch, and applies reversible retention scores to cut bridging edges; The determination unit is used to determine the node type based on the edge weight, bridging entropy, significant connection number, and attribution weight; and to determine the cluster state based on the cluster's stability, activity, and dispersion. The output unit is used to fuse multi-source evidence, adjust the evidence propagation intensity according to node type and cluster state, fuse it with historical membership after bidirectional propagation, and calculate an adaptive threshold based on node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

[0018] Thirdly, the present invention also provides a computer-readable storage medium storing computer instructions that cause a computer to perform an overlapping clustering method based on a bipartite graph batch Chinese restaurant process as described in the first aspect or any embodiment of the first aspect.

[0019] The embodiments of the present invention bring the following beneficial effects: This invention proposes an overlapping clustering method based on a batch processing of restaurants using a bipartite graph. This method is designed for incremental data batches arriving sequentially over time. Without revisiting all historical samples, it first adaptively generates candidate clusters. Second, using the batch processing of restaurants, it performs a joint allocation of samples, incorporating existing clusters, triggering new clusters, and retaining overlapping candidates. Third, it constructs a dynamic bipartite graph and uses locally reversible constraints to cut and maintain the overlapping boundary structure. Finally, it outputs the final overlapping clustering result through evidence propagation and fusion of overlapping structures. This method inherits historical cluster representations and edge weight memories, eliminating the need to replay historical samples. It achieves dynamic modeling of cluster evolution behavior and multi-cluster membership while maintaining computational efficiency, making it suitable for data analysis scenarios with multiple semantic overlaps, such as multi-label learning, social network analysis, and bioinformatics.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 A flowchart of an overlapping clustering method for batch Chinese restaurant processing based on bipartite graphs provided in an embodiment of the present invention; Figure 2 A flowchart of an algorithm based on a restaurant process overlap clustering model in a bipartite graph, provided for another embodiment of this application; Figure 3 A detailed framework diagram of the overlapping clustering process provided for another embodiment of the present invention; Figure 4 The structural block diagram of the overlapping clustering system for batch Chinese restaurant processing based on bipartite graphs provided in this application embodiment is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0025] Current technologies for dynamic overlapping clustering still have shortcomings: First, most existing overlapping clustering methods are based on static batch processing, making it difficult to adapt to the dynamic evolution of cluster structures under incremental data conditions. Second, while existing dynamic clustering methods can handle data stream updates, they often struggle to effectively characterize the overlapping membership relationships of multiple clusters. Third, although existing bipartite graph clustering methods can utilize graph structure information for sample partitioning, they generally lack the ability to integrate with dynamic incremental learning mechanisms, making it difficult to achieve unified processing of dynamic bipartite graph construction, graph cutting, key node selection, and subsequent overlapping cluster generation. Fourth, some methods still rely on manually preset cluster numbers, empirical thresholds, or computationally expensive inference mechanisms, making it difficult to meet the application needs of complex dynamic scenarios. Therefore, it is necessary to propose a new clustering construction method to address the difficulty in simultaneously achieving dynamism, overlap, and efficiency in existing technologies.

[0026] Based on this, the embodiments of the present invention provide an overlapping clustering method, system, and medium based on a batch Chinese restaurant process using a bipartite graph. This method can construct a dynamic bipartite graph between sample nodes and cluster nodes, and introduce a batch Chinese restaurant process to achieve incremental clustering of samples and adaptive generation of new clusters. Combined with graph cutting, key node selection, and label propagation and fusion mechanisms, it can realize the dynamic construction, updating, and expression of overlapping cluster structures. This improves the efficiency of incremental processing while enhancing the ability to model overlapping relationships and cluster evolution behavior in complex dynamic data.

[0027] To facilitate understanding of this embodiment, a detailed description of an overlapping clustering method for batch Chinese restaurant processing based on bipartite graphs, as disclosed in this embodiment of the invention, will be provided first.

[0028] Figure 1 This is a flowchart illustrating an overlapping clustering method for batch processing of Chinese restaurants based on a bipartite graph, provided in an embodiment of the present invention. Figure 1 As shown, the method may include: S102: Receive incremental data from the current batch, and generate candidate new clusters based on the compatibility, novelty, and batch drift of the samples in the incremental data with historical clusters.

[0029] It should be noted that the embodiments of the present invention are applicable to data analysis scenarios where dynamism, overlap, and incrementality coexist. They can be used for multi-label learning, discovery of overlapping communities in social networks, multifunctional gene clustering in bioinformatics, identification of multi-interest groups in recommendation systems, and dynamic overlapping clustering analysis of multi-source heterogeneous data such as text, images, and logs.

[0030] In this embodiment, given a high-dimensional incremental data stream arriving in chronological order. in, For the first Batch sample set, This represents the current batch sample size. For sample dimensions. In this embodiment, samples are allowed to have multiple overlapping memberships, using a soft membership matrix. Indicates the relationship between samples and clusters. For a moment The number of clusters, Indicates sample Cluster The strength of association. Construct a dynamic bipartite graph consisting of sample nodes and cluster nodes:

[0031] in, For the set of sample nodes, For a set of cluster nodes, Let be the set of edges. This is the edge weight matrix. The cluster center is maintained for each cluster node. Cluster discrete structure Cluster size Cluster stability Cluster activity and cluster evolution state .

[0032] In batch Upon arrival, the number of candidate new clusters for the current batch is adaptively determined based on the compatibility relationship between the current batch of samples and historical clusters, their own novelty, and the evolutionary state of historical clusters. For samples... With historical clusters The sample-cluster compatibility is defined as:

[0033] in, This is a consistency term between the sample and cluster structure distribution. For the compatibility of the sample with the historical label structure, , , These are non-negative weighting coefficients. Let be the center of cluster k after the previous batch, and ε be a very small positive number. Therefore, sample novelty is defined as follows:

[0034] in, For the uncertainty of attribution, This represents the local outlier score. Further, the average novelty score of the current batch is defined as follows:

[0035] Define the batch drift of the current batch relative to the historical structure:

[0036] in, This is the mean of the current batch of samples. Weighted mean of historical cluster centers; redefining This represents the average splitting risk intensity at the overall historical cluster level, used to adjust the number of new candidate clusters added in the current batch during batch initialization:

[0037] Based on the above quantities, an adaptive concentration parameter is defined:

[0038] This allows for a greater likelihood of generating new clusters when the current batch is newer, deviates more from the historical structure, and is closer to the cluster fission state; conversely, the more stable the historical clusters, the more ineffective cluster generation is suppressed. The number of candidate new clusters is determined based on the Poisson decay prior.

[0039] in, The time decay exponent, To initialize the strength, for each candidate new cluster, select samples with high novelty and local support from the current batch as initial centers. Local support is defined as:

[0040] Only when the sample satisfies and Only when a noise point is identified can it be considered as a candidate new cluster center, thus avoiding the accidental triggering of new clusters by isolated noise points.

[0041] S104: Calculate the scores of the samples assigned to existing clusters and new clusters using the batch Chinese restaurant process, and retain overlapping candidate clusters based on the scores.

[0042] After initializing the candidate new clusters, perform batch restaurant allocation on the current batch of samples. For any sample and First, define intra-batch similarity:

[0043] in, For feature similarity, For neighborhood structural similarity, These are non-negative weighting coefficients. (Sample) Assigned to an existing cluster The first score is defined as:

[0044] in, For batch-intra-batch collaborative items, it is represented as:

[0045] Used for characterization and samples Do highly similar samples in the same batch jointly support clusters? ; For cross-batch memory items, it is represented as:

[0046] Used to introduce the continuity constraint of historical neighbors on the current sample's attribution; This is a cluster state mapping function, which assigns different adjustment values ​​based on whether the cluster is in an active, stable, drifting, splitting, or merging state, thus allowing the cluster evolution state to directly affect the current batch allocation. (Sample) The second score for generating new clusters is defined as:

[0047] That is, when both sample novelty and local microcluster support are high, there is a greater tendency to trigger new clusters. The posterior assignment probability is obtained by normalizing the scores of existing and new clusters:

[0048]

[0049] Unlike traditional unique assignment, the overlapping candidate cluster set of samples preserved in this embodiment of the invention is represented as follows:

[0050] The initial soft membership matrix for the current batch is defined as follows:

[0051] Therefore, the multi-cluster overlap relationship of the samples is formed directly in the clustering stage, rather than being compensated and generated after the clustering is completed.

[0052] S106: Construct a dynamic bipartite graph and edge weights for nodes and cluster nodes, calculate bridging entropy based on edge weights to identify bridging edges; perform cutting on local subgraphs affected by the current batch, and apply reversible retention scores to cut bridging edges.

[0053] After completing the batch allocation of Chinese restaurants, a dynamic bipartite graph is constructed using sample nodes and cluster nodes. If... Then at the sample node With cluster nodes Establish edges between Edge weight is defined as:

[0054] right , , and Normalize them separately so that their values ​​all fall within the range The interval. Wherein, 、 、 and These are non-negative weighting coefficients; For the current attribution strength, For sample-cluster compatibility, The historical connection memory item is represented as:

[0055] in, The average connection strength over a historical window. The time interval since the most recent high-strength connection; For bridging and holding terms, it is represented as:

[0056] Used to preserve stable bridging relationships in historical data. Based on edge weight distribution, edges are divided into core edges, bridging edges, and drift edges: when a sample has a significant primary affiliation to a cluster, the corresponding edge is marked as a core edge; when a sample maintains high connectivity with multiple clusters simultaneously, the corresponding edge is marked as a bridging edge; when the current affiliation has high support but low historical compatibility, the corresponding edge is marked as a drift edge. Cluster node parameters are updated synchronously, with the cluster center updated as follows:

[0057] Cluster size updated to:

[0058] Cluster stability updated to:

[0059] Cluster activity updated to:

[0060] Then, based on cluster stability, activity, and cross-cluster overlap, the cluster evolution state is updated. This enables dynamic management of cluster generation, expansion, drift, splitting, and merging.

[0061] To avoid disrupting the overlapping boundary structure during conventional graph cutting, this embodiment of the invention does not perform recutting on the global bipartite graph, but only on the local subgraphs affected by the current batch. Perform constraint-based cutting. First, calculate the normalized connection weights for the sample nodes:

[0062] And further define the sample bridging entropy:

[0063] When a sample is connected to multiple clusters with similar connection strengths, the bridging entropy is high, indicating that the sample is located in an overlapping boundary region. Define the set of bridging edges:

[0064] Define the basic partition label of the sample node Whether to cut the edge is indicated as Then the locally reversible cutting target is:

[0065] in To retain the fraction for reversibility, it is represented as:

[0066] This is a time-series smoothing term used to suppress drastic fluctuations in the cutting results of adjacent batches. If a bridging edge is cut in the current batch but not in subsequent batches... In each batch, the following conditions are met If the edge connection is automatically restored, a locally reversible cutting mechanism is formed, which reduces the cost of reconstruction while avoiding the accidental cutting of overlapping bridging relationships.

[0067] S108: Determine the node type based on edge weight, bridging entropy, significant connection number, and affiliation weight; determine the cluster state based on cluster stability, activity, and dispersion.

[0068] Core nodes are identified using the maximum edge weight and bridging entropy, bridging nodes are identified using bridging entropy and significant connection number, and boundary nodes are identified using the difference between the maximum and second-largest attribution weights.

[0069] After the locally reversible constraint cutting is completed, core nodes, bridging nodes, and boundary nodes are identified from the sample nodes for subsequent overlap propagation and cluster evolution control. The coreness of the sample nodes is defined as:

[0070] when A high bridging degree indicates that the sample has a significant primary affiliation to a certain cluster and weak bridging ability; therefore, it is used as the core seed node. The bridging degree is defined as:

[0071] when A higher value indicates that the sample connects to multiple clusters simultaneously, serving as a key bridging node expressing overlapping relationships. If the difference between the sample's largest and second-largest affiliation weights is less than a threshold... If the sample is not found, it is designated as a boundary node. Based on the distribution of core nodes, bridging nodes, and boundary nodes, and combined with cluster stability, cluster activity, and cluster dispersion, cluster evolution decisions are made: when cluster activity continues to rise and stability is high, it remains an active cluster; when cluster dispersion continues to increase and stability decreases, it is marked as a cluster to be split; when the proportion of shared samples between two clusters is high and the center distance is below a threshold, cluster merging is performed; when cluster activity is below a threshold for a long period, it is marked as a dormant cluster.

[0072] S110: Integrates multi-source evidence, adjusts the evidence propagation intensity according to node type and cluster state, integrates it with historical membership after bidirectional propagation, and calculates an adaptive threshold based on node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

[0073] After the key nodes are identified, the sample-cluster relationship is updated bidirectionally using the overlapping structure evidence propagation mechanism, and historical label memory is integrated to output the final overlapping clustering result.

[0074] Define sample Belongs to cluster The cumulative evidence is as follows:

[0075] The first term represents the current posterior assignment evidence, the second term represents the current graph connectivity evidence, the third term represents historical attribution evidence, and the fourth term represents intra-batch collaboration evidence. Let the initial evidence matrix on the sample side be... The initial evidence matrix on the cluster side is Then the bidirectional propagation update between sample nodes and cluster nodes is as follows:

[0076]

[0077] When satisfied Stop iteration when the propagation ends. After propagation is complete, define:

[0078] This is to prevent the unbounded spread of evidence during the propagation process. Then, the current propagation results are integrated with historical results:

[0079] in This is the historical tag mapping function. The normalized bridging entropy is defined as:

[0080] Therefore, the adaptive threshold is:

[0081] The final overlapping cluster set is defined as

[0082] The final output includes the set of overlapping clusters, sample-cluster soft membership matrix, dynamic bipartite graph structure, set of key nodes, and cluster generation, expansion, splitting, and merging states of the current batch of samples.

[0083] This invention employs a mechanism combining Poisson decay initialization with cluster evolution state coupling, batch processing, dynamic bipartite graph update, and overlapping structure evidence propagation to construct overlapping clusters for the current batch of samples without revisiting all historical samples. By introducing historical cluster stability, cross-batch memory terms, locally reversible cuts, and smoothing constraints, it can suppress drastic oscillations in cluster structure and maintain the temporal continuity and update stability of clustering results under continuous incremental data arrival. Let the number of samples in the current batch be... The current number of clusters is The local neighborhood size is The number of sides in the dynamic graph is The number of propagation iterations is The time complexity of a single batch is approximately The space complexity is approximately .

[0084] As an optional implementation, the process of generating candidate new clusters includes: Determine whether a new pattern not seen in the history appears in the current batch of samples. If so, determine the number of new clusters based on the compatibility of the samples with historical clusters, sample novelty, overall batch drift, and risk of historical cluster splitting. Initialize the new cluster centers from samples that meet the novelty threshold and local support.

[0085] As an optional implementation, a batch processing procedure for Chinese restaurants is used to calculate scores for assigning samples to existing and new clusters, and overlapping candidate clusters are retained based on these scores, including: Intra-batch similarity is calculated based on feature similarity and neighborhood structure similarity between samples; The first score for assigning a sample to each existing cluster and the second score for assigning a sample to each candidate new cluster are calculated. The score for assigning a sample to an existing cluster incorporates the historical size of the cluster, the compatibility between the sample and the cluster, intra-batch collaboration, cross-batch memory, and cluster state mapping adjustment. The score for assigning a sample to a new cluster incorporates the adaptive concentration parameter, the sample novelty, and the local support of the sample to the candidate new cluster. Normalize the first and second scores to obtain the probability that the sample belongs to each cluster; Based on the relative probability threshold, overlapping candidate clusters are selected and retained to determine the multi-cluster overlapping membership relationship of the samples.

[0086] As an optional implementation, a dynamic bipartite graph of nodes and cluster nodes and edge weights is constructed, including: Edge weights are determined by a combination of the current node's affiliation strength, sample-cluster compatibility, historical connection memory, and bridging preservation term. Edge states are classified according to their weight distribution, including core edges, bridging edges, and drifting edges.

[0087] As an optional implementation, a slicing process is performed on the local subgraph affected by the current batch, including: Based on the bridging entropy of the samples, identify the samples and bridging edges located in the overlapping boundary region; When a bridging edge is cut, an additional cost is imposed on the bridging edge based on the historical edge weight, cluster stability, and current edge weight; if the cut bridging edge satisfies the reversibility condition in subsequent consecutive batches, the connection of the bridging edge is restored.

[0088] As an alternative implementation, the reversible retention score is obtained by weighted fusion of historical edge weights, cluster stability, and current edge weights.

[0089] As an optional implementation, the process of determining the cluster state includes: Clusters whose activity increases in consecutive batches and whose cluster center changes are below a threshold are kept as active clusters; clusters whose sample dispersion increases in consecutive batches and whose cluster center changes are above a threshold are marked as clusters to be split; two clusters whose shared sample ratio is above a threshold and whose cluster center distance is below a threshold are merged; clusters whose activity is below a threshold in consecutive batches are marked as dormant clusters.

[0090] As an optional implementation, an adaptive threshold is calculated based on node type, bridging entropy, and cluster state, including: The threshold for core nodes is higher than that for bridging nodes and boundary nodes; the threshold for sample nodes belonging to the cluster to be split or merged is lower than that for sample nodes belonging to the active or dormant cluster; when a sample is both a core node and belongs to the cluster to be split, the lower threshold corresponding to the cluster to be split is used.

[0091] Figure 2 A flowchart of an algorithm based on a restaurant process overlap clustering model in a bipartite graph, provided for another embodiment of this application; Figure 3 A detailed framework diagram of an overlapping clustering process provided in another embodiment of the present invention is shown below. The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The present invention includes, but is not limited to, the following embodiments. Figure 2 As shown, taking bioinformatics applications as an example, the specific implementation process is as follows: In bioinformatics analysis, the same gene or protein often participates in multiple biological processes or signaling pathways simultaneously. For example, some proteins are involved in both immune responses and cell proliferation regulation, thus exhibiting overlapping affiliation characteristics across multiple clusters. Furthermore, transcriptome sequencing, proteome detection, and multi-omics joint analysis typically generate new samples continuously in batches. Re-clustering all historical samples each time would incur high computational costs and potentially disrupt existing stable cluster structures. Therefore, this invention is particularly suitable for constructing multifunctional overlapping clusters under dynamic incremental sample conditions in bioinformatics.

[0092] This invention uses dynamic identification of gene functional modules as an example. Assuming the samples are candidate genes to be analyzed, each gene is represented by a 4-dimensional feature vector, corresponding to: normalized expression intensity under inflammatory stimulation; cell cycle-related expression activity; expression response under metabolic stress; and local connectivity strength in the protein-protein interaction network. For ease of explanation, two batches of incremental data, each containing five gene samples, are selected for demonstration. In practical applications, the above feature dimensions can be expanded to higher-dimensional transcriptomic features, proteomic features, domain features, pathway scoring features, or multi-omics fusion representations.

[0093] The main parameters used in this embodiment can be set as follows: , , , , , , , , , , , , , The aforementioned sample-cluster compatibility, intra-batch similarity, bridging entropy, adaptive concentration parameter, and propagation process are all calculated according to the definitions in the technical solution of this invention.

[0094] (1) Input, calculation and output of the first batch of gene samples The first batch of input consisted of 5 candidate gene samples, denoted as follows: Its 4-dimensional features are represented as follows:

[0095]

[0096]

[0097]

[0098]

[0099] Among them, samples and It exhibits high expression activity in inflammatory responses and can be considered a potential immune response-related gene; sample and It exhibits high cell cycle-related activity and can be considered a potential cell proliferation-related gene; the sample It also exhibits medium to high levels of response in inflammatory responses and cell cycle dimensions, and is a potential multifunctional overlapping gene.

[0100] When the first batch of data arrived, there were no historical clusters in the system, i.e. Therefore, based on the cluster evolution state coupled with Poisson decay initialization mechanism of S1 in this invention, the average novelty of the current batch is high, and there are no historically stable cluster suppression terms. The calculated number of candidate new clusters for the current batch is 2. Further combining sample novelty and local support, clusters are selected from the current batch. and These two clusters will serve as the initial centers for the two candidate new clusters, denoted as clusters. and cluster .

[0101] Subsequently, the competitive-cooperative batch allocation of Chinese restaurants in S2 is performed. Based on intra-batch similarity and sample-cluster compatibility calculations, the samples... and Cluster The posterior belonging probability is significantly higher than that of the cluster. ;sample and Cluster The posterior attribution probability is significantly higher; the sample The posterior membership probabilities on the two clusters are similar. After normalization, the soft membership matrix of the current batch can be approximated as:

[0102] Based on the overlapping candidate retention threshold ,sample Simultaneously retain clusters and cluster The candidate attributions are determined, thus obtaining the overlapping label matrix for the current batch:

[0103] Among them, the 5th line Indicates gene sample It belongs to both the immune response function cluster and the cell proliferation function cluster, reflecting that it may be involved in both inflammation regulation and cell cycle regulation.

[0104] Next, execute S3 to construct a dynamic bipartite graph of sample nodes and cluster nodes. Among them, the sample and With cluster nodes Forming core edges; samples and With cluster nodes Forming core edges; samples A bridging edge is formed between the two cluster nodes. Based on the edge weight update rule, the approximate centers of the two clusters are:

[0105] Further perform the locally reversible constraint cut of S4. Due to the sample... Cluster and cluster The normalized edge weights are close, and the bridging entropy is large. Therefore, the connecting edges are identified as bridging edges and retained, without hard cutting. This forms two basic functional clusters, while retaining the intermediate overlapping boundaries.

[0106] In S5, samples Identified as a cluster Core nodes, samples Identified as a cluster Core nodes, samples The node was identified as a bridging node. Finally, S6's overlap evidence conservation propagation and fusion were performed. Since the first batch had no historical labels, the propagation result was mainly determined by the posterior assignment evidence, graph connectivity evidence, and intra-batch cooperation evidence of the current batch. After propagation, the final overlap cluster result was obtained:

[0107] Therefore, after the first batch of treatment, an immune response functional cluster was obtained. A cluster of cell proliferation functions and an overlapping gene sample belonging to both. The system simultaneously outputs a label matrix. Cluster statistics and dynamic bipartite graph structure This will be used as historical information for the next batch of data processing.

[0108] (2) Input, calculation and output of the second batch of gene samples The second batch of input consisted of 5 candidate gene samples, denoted as follows: Its 4-dimensional features are represented as follows:

[0109]

[0110]

[0111]

[0112]

[0113] Among them, samples The immune response clusters are quite similar to those in the previous batch; samples and Similar to cell proliferation clusters; sample It showed a significant enhancement in the metabolic stress dimension, which may correspond to a new functional cluster of metabolic stress; Sample It also exhibits high responsiveness in the cell cycle and metabolic stress dimensions, and belongs to the potential overlapping boundary genes.

[0114] When the second batch of data arrives, the system has already saved the historical cluster structure obtained in the first batch. After calculating sample novelty, overall drift, and global split risk based on S1, it was found that the sample... For existing clusters and The compatibility is low, but it is compatible with the sample. There is a certain support relationship in the local neighborhood, therefore the adaptive concentration parameter of the current batch The number has increased compared to the previous batch. Based on the Poisson decay initialization results, the system generates one new candidate cluster, denoted as [Cluster Name]. Its initial center is taken as the sample .

[0115] Then, the competitive-cooperative batch processing of Chinese restaurants in S2 is executed. Based on sample-cluster compatibility and intra-batch collaboration terms, the sample... Mainly belongs to cluster ;sample and Mainly belongs to cluster ;sample Mainly belongs to the newly added cluster ;sample Simultaneously for clusters and cluster It has a high posterior affiliation. After normalization, the soft membership matrix of the current batch can be approximated as:

[0116] This yields the overlapping label matrix for the current batch:

[0117] Among them, the 5th line Indicates sample It also belongs to the cell proliferation function cluster. and metabolic stress function cluster This suggests that the gene may be involved in both cellular stress reprogramming and proliferation regulation.

[0118] Next, execute S3 to build and update the dynamic bipartite graph. Among them, the sample With cluster nodes Forming core edges, samples and With cluster nodes Forming core edges, samples With newly added cluster nodes Forming core edges, samples At the same time with and Establish bridging edges. After updating, the centers of the three clusters are approximately:

[0119]

[0120]

[0121] Further perform the locally reversible constraint cut of S4. Sample In cluster and cluster The normalized edge weights between clusters are close, resulting in a large bridging entropy. Therefore, these bridging edges are preserved without forced severing; the remaining low-weight and non-critical cross-cluster edges are removed. This creates new metabolic stress clusters while preserving the potential functional crossover between cell proliferation and metabolic stress.

[0122] In S5, samples Identified as a cluster Core nodes, samples Identified as a cluster Core nodes, samples Identified as a cluster Core nodes, samples It was identified as a bridging node. Based on cluster activity and stability analysis, the system determined the cluster... and cluster Stay active, cluster As a newly generated cluster, it is in an active state.

[0123] Finally, the overlapping evidence conservation propagation and history fusion in step S6 are performed. Because this step introduces the first batch of historical label information, the final assignment of samples in the current batch is influenced not only by the current posterior assignment evidence and graph connectivity evidence, but also by the cluster structure memory of the previous batch. Therefore, it can suppress cluster label mutations caused by incremental arrivals. After propagation, the final overlapping cluster results are obtained:

[0124] Therefore, after the second batch of processing is completed, the system ultimately consists of three functional modules, clusters. Indicates functional modules related to immune response; cluster Represents functional modules related to cell proliferation; clusters This indicates functional modules related to metabolic stress. Meanwhile, the sample... The sample was identified as an overlapping gene between immune response and cell proliferation. It was identified as a cell proliferation-metabolic stress overlap gene, reflecting the real biological phenomenon that the same gene can participate in multiple functional processes simultaneously in an organism.

[0125] (3) Hyperparameter settings The parameters involved in this embodiment include the initial concentration parameter. Novelty weight Drift weight Split risk weight Stability suppression weights Overlapping candidate retention threshold Bridge edge identification threshold Bridging entropy threshold Propagation convergence threshold and the upper and lower bounds of the output threshold and The specific values ​​of the above parameters will affect the sensitivity of new functional cluster generation, the degree of overlap boundary preservation, and the final multifunctional membership output results.

[0126] for and Its main function is to adjust the initialization capability of candidate new clusters in the current batch. Larger... and This will enhance the model's sensitivity to new gene functional patterns and batch drift, facilitating the rapid discovery of new biological process modules; larger This will enhance the model's ability to inherit existing stable functional modules, which is beneficial for suppressing excessive splitting and ineffective clustering. For It determines whether a sample retains multiple candidate functional clusters, with larger ones being more likely to be included. It tends to obtain sparser overlapping memberships, smaller This is beneficial for preserving richer, more multifunctional semantics. For and This determines the identification thresholds for bridging edges and nodes, thus affecting the level of protection for overlapping boundaries during the dynamic bipartite graph cutting stage. For and This determines the ability of the final output stage to distinguish between bridging samples and core samples.

[0127] This embodiment illustrates how, in a bioinformatics application scenario, the present invention achieves dynamic overlapping clustering construction and updating of gene functional modules through a complete process including candidate new cluster initialization, collaborative allocation, dynamic bipartite graph construction, local reversible cutting, key node identification, and overlapping evidence propagation and fusion. It should be noted that although most details of the present invention have been described in detail using the above examples, this does not limit the present invention to the specific embodiments described. Furthermore, adjustments or modifications can still be made to the above technical solutions based on the content of this specification. Any modifications or equivalent substitutions made based on the technical solutions of this invention should fall within the scope of protection claimed by the present invention.

[0128] Figure 4 The structural block diagram of the overlapping clustering system for batch Chinese restaurant processing based on bipartite graphs provided in this application embodiment is as follows: Figure 4 As shown, the system may include: The generation unit 401 is used to receive incremental data of the current batch and generate candidate new clusters based on the compatibility, novelty and batch drift of the samples in the incremental data with historical clusters. Calculation unit 403 is used to calculate the scores of samples assigned to existing clusters and new clusters respectively using the batch Chinese restaurant process, and retain overlapping candidate clusters based on the scores; Cutting unit 405 is used to construct a dynamic bipartite graph of nodes and cluster nodes and edge weights, calculate bridging entropy based on edge weights to identify bridging edges, perform cutting on local subgraphs affected by the current batch, and apply reversible retention scores to cut bridging edges; The determination unit 407 is used to determine the node type based on the edge weight, bridging entropy, significant connection number and belonging weight; and to determine the cluster state based on the cluster's stability, activity and dispersion. Output unit 409 is used to fuse multi-source evidence, adjust the evidence propagation intensity according to node type and cluster state, fuse it with historical membership degree after bidirectional propagation, and calculate adaptive threshold according to node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

[0129] The system provided in this application embodiment has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0130] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0131] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0133] This application provides a computer-readable storage medium that stores computer instructions, which cause the computer to execute embodiments of this application. Figure 1 The method provided in the illustrated embodiment.

[0134] The aforementioned computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0135] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0136] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0137] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0138] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0139] In the description of the embodiments of this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In the embodiments of this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of this application, as well as the features of different embodiments or examples.

[0140] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0141] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0142] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0143] It should be noted that the terminals involved in the embodiments of this application may include, but are not limited to, personal computers (PCs), personal digital assistants (PDAs), wireless handheld devices, tablet computers, mobile phones, MP3 players, MP4 players, etc.

[0144] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0145] Furthermore, the functional units in the various embodiments of this application can be integrated into a single processor, or each unit can exist physically separately, or two or more units can be integrated into a single unit. The integrated units described above can be implemented in hardware or in a combination of hardware and software functional units.

[0146] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present application should be included within the scope of protection of the present application.

Claims

1. An overlapping clustering method based on a bipartite graph for batch processing of Chinese restaurants, characterized in that, include: Receive incremental data from the current batch, and generate candidate new clusters based on the compatibility, novelty, and batch drift of the samples in the incremental data with historical clusters; The scores of the samples assigned to existing clusters and new clusters are calculated using a batch Chinese restaurant process, and overlapping candidate clusters are retained based on the scores; Construct a dynamic bipartite graph of nodes and cluster nodes and edge weights, calculate bridging entropy based on the edge weights to identify bridging edges; perform cutting on local subgraphs affected by the current batch, and apply reversible retention scores to the cut bridging edges; The node type is determined based on the edge weight, bridging entropy, significant connection number, and attribution weight. The cluster state is determined based on the cluster's stability, activity, and dispersion. Multi-source evidence is fused, and the evidence propagation intensity is adjusted according to the node type and cluster state. After bidirectional propagation, it is fused with historical membership degree, and an adaptive threshold is calculated based on the node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

2. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The process of generating candidate new clusters includes: Determine whether a new pattern not seen in the history appears in the current batch of samples. If so, determine the number of new clusters based on the compatibility of the samples with historical clusters, sample novelty, overall batch drift, and risk of historical cluster splitting. Initialize the new cluster centers from samples that meet the novelty threshold and local support.

3. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The process of calculating the scores of the samples assigned to existing and new clusters using a batch Chinese restaurant process, and retaining overlapping candidate clusters based on the scores, includes: Intra-batch similarity is calculated based on feature similarity and neighborhood structure similarity between samples; The first score of the sample assigned to each existing cluster is calculated, and the second score of the sample assigned to each candidate new cluster is calculated; wherein, the score assigned to an existing cluster integrates the historical size of the cluster, the compatibility between the sample and the cluster, the intra-batch collaboration term, the cross-batch memory term, and the cluster state mapping adjustment amount; the score assigned to a new cluster integrates the adaptive concentration parameter, the sample novelty, and the local support of the sample to the candidate new cluster. Normalize the first score and the second score to obtain the probability that the sample belongs to each cluster; Based on the relative probability threshold, overlapping candidate clusters are selected and retained to determine the multi-cluster overlapping membership relationship of the samples.

4. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The construction of the dynamic bipartite graph and edge weights of the nodes and cluster nodes includes: The edge weights are determined by a combination of the current node's affiliation strength, sample-cluster compatibility, historical connection memory, and bridging preservation term. The edge states are classified according to the edge weight distribution, including: core edges, bridging edges, and drifting edges.

5. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The process of cutting the local sub-graph affected by the current batch includes: Based on the bridging entropy of the samples, identify the samples and bridging edges located in the overlapping boundary region; When the bridging edge is cut, an additional cost is imposed on the bridging edge based on the historical edge weight, cluster stability, and current edge weight; if the cut bridging edge satisfies the reversibility condition in subsequent consecutive batches, the connection of the bridging edge is restored.

6. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 5, characterized in that, The reversible retention score is obtained by weighted fusion of historical edge weights, cluster stability, and current edge weights.

7. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The process of determining the cluster state includes: Clusters whose activity increases in consecutive batches and whose cluster center changes are below a threshold are kept as active clusters; clusters whose sample dispersion increases in consecutive batches and whose cluster center changes are above a threshold are marked as clusters to be split; two clusters whose shared sample ratio is above a threshold and whose cluster center distance is below a threshold are merged; clusters whose activity is below a threshold in consecutive batches are marked as dormant clusters.

8. The overlapping clustering method for batch processing of Chinese restaurants based on bipartite graphs according to claim 1, characterized in that, The calculation of the adaptive threshold based on the node type, bridging entropy, and cluster state includes: The threshold for core nodes is higher than that for bridging nodes and boundary nodes; the threshold for sample nodes belonging to the cluster to be split or merged is lower than that for sample nodes belonging to the active cluster or dormant cluster; when a sample is both a core node and belongs to the cluster to be split, the lower threshold corresponding to the cluster to be split is used.

9. An overlapping clustering system based on a bipartite graph-based batch processing of Chinese restaurants, characterized in that, include: The generation unit is used to receive incremental data of the current batch and generate candidate new clusters based on the compatibility, novelty and batch drift of the samples in the incremental data with historical clusters. The calculation unit is used to calculate the scores of the samples assigned to existing clusters and new clusters respectively using the batch Chinese restaurant process, and retain overlapping candidate clusters based on the scores; The cutting unit is used to construct a dynamic bipartite graph of nodes and cluster nodes and edge weights, calculate bridging entropy based on the edge weights to identify bridging edges, perform cutting on local subgraphs affected by the current batch, and apply reversible retention scores to the cut bridging edges; The determination unit is used to determine the node type based on the edge weight, bridging entropy, significant connection number, and home weight. The cluster state is determined based on the cluster's stability, activity, and dispersion. The output unit is used to fuse multi-source evidence, adjust the evidence propagation intensity according to the node type and cluster state, fuse it with historical membership degree after bidirectional propagation, and calculate an adaptive threshold according to the node type, bridging entropy and cluster state to output the overlapping cluster set of each sample.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to perform the overlapping clustering method based on the bipartite graph batch Chinese restaurant process as described in any one of claims 1 to 8.

Images