Semi-supervised multi-view clustering method based on concept decomposition of double constraints

Abstract

The application discloses a semi-supervised multi-view clustering method based on double-constraint concept decomposition, relates to the technical field of machine learning, combines point constraints with pair constraints, constructs a double-constraint algorithm based on prior label information of a complementary supervision mechanism, realizes maximum utilization and progressive propagation of limited supervision information, breaks through the limitation of data non-negativity of a traditional method by relying on a concept decomposition framework, processes complex data distribution by combining a kernel method, obtains a consensus matrix of multi-view data fusion as low-dimensional representation of a view, adopts a k-means algorithm for clustering, and outputs a final class division result. The application realizes effective processing of complex multi-view data, finally comprehensively improves the performance of a multi-view clustering task, and breaks through the bottleneck of the prior art in clustering precision, data adaptability and supervision information utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and more specifically, to a semi-supervised multi-view clustering method based on dual-constraint concept decomposition. Background Technology

[0002] With the widespread availability of multimodal data, multi-view clustering has become a research hotspot due to its ability to integrate complementary information from different sources. While nonnegative matrix factorization (NMF) methods are widely adopted, they are limited by nonnegative data processing and struggle to capture nonlinear structures. Concept factorization (CF), as a generalization of NMF, overcomes the nonnegativity limitation of data and is compatible with kernel methods, becoming a new direction for multi-view clustering (e.g., MVCC, MVDGCF models). However, most existing CF methods are unsupervised learning, failing to effectively utilize the limited labeling information commonly found in real-world scenarios. Although some semi-supervised methods have attempted this approach, they still suffer from insufficient utilization of supervised information and limited constraint forms, resulting in limited performance improvements in sparsely labeled scenarios.

[0003] Existing multi-view clustering methods have significant shortcomings in improving clustering performance by utilizing limited supervised information. Specifically, most concept decomposition (CF)-based multi-view clustering methods employ unsupervised learning strategies, failing to effectively utilize the limited label information commonly found in real-world data, thus limiting clustering accuracy. A few semi-supervised multi-view CF methods utilize supervised information in a simplistic way, processing labeled data only through simple methods like label propagation, failing to fully explore the potential value of labeled information. Furthermore, existing methods lack complementary mechanisms for utilizing supervised information; they neither directly guarantee clustering consistency for samples with known categories through point constraints nor effectively propagate labeled information to unlabeled samples through pairwise constraints. This results in limited coverage of supervised information, making it difficult to guide the overall clustering structure towards the true distribution. These problems limit the performance improvement of existing methods in sparsely labeled scenarios, failing to meet the accuracy and robustness requirements of multi-view data clustering. Therefore, overcoming the shortcomings of existing multi-view clustering techniques in terms of supervised information utilization and data processing, and providing an efficient semi-supervised multi-view clustering scheme, is an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a semi-supervised multi-view clustering method based on dual-constraint concept decomposition. This method addresses the performance limitations of semi-supervised multi-view clustering in scenarios with scarce labeling information by maximizing the utilization and gradual diffusion of limited supervised information, fully leveraging a small amount of prior information, and improving clustering accuracy in scenarios with low labeling rates.

[0005] The first aspect of this invention provides a semi-supervised multi-view clustering method based on dual-constraint concept decomposition, comprising the following steps:

[0006] Obtain a multi-view dataset containing multiple view samples, some of which have known labels. Initialize the multi-view shared label constraint matrix, pairwise constraint matrix, representation matrix, association matrix, and consensus matrix in the multi-view dataset.

[0007] For each view, an initial weight matrix is ​​constructed based on nearest neighbor relationships and kernel functions. The final pairwise constraint matrix is ​​generated through vertical and horizontal constraint propagation algorithms. The weight matrix is ​​updated to enhance the correlation of samples of the same type. The corresponding graph Laplacian matrix and diagonal matrix are calculated.

[0008] In the alternating iteration, the projection matrix is ​​fixed and the association matrix is ​​updated using the shared label constraint matrix. The association matrix is ​​fixed and the projection matrix is ​​updated using the graph Laplacian matrix and the diagonal matrix. After each iteration, the consensus matrix is ​​updated based on the projection matrix of each view.

[0009] After the iteration is completed, the final consensus matrix is ​​used as a low-dimensional representation of the view, and the k-means algorithm is used for clustering to output the final category division results.

[0010] In this scheme, a multi-view dataset containing sample data from multiple views is obtained, where some samples have known labels. The shared label constraint matrix, representation matrix, projection matrix, and consensus matrix of the view sample data in the multi-view dataset are initialized as follows:

[0011] Input multi-view dataset X = [X (1) ,X (2) ,…,X (v) ,…,X (nv) ], where X (v) Let v represent the data matrix of the v-th view. The multi-view dataset contains n views, each view has m feature dimensions, the first p samples have known class labels, and the remaining np samples have no labels. The number of classes is c.

[0012] Encoding known category label information C for samples with known category labels p×c Using the identity matrix I n-p This represents the initial free representation of an unknown-labeled sample, encoded based on the known label portions. p×c The unknown label part indicates I n-p Construct a multi-view shared label constraint matrix

[0013] Based on samples with known category labels, pairwise constraints are constructed. The constraint relationship is constructed when the samples in a sample pair are of the same type. The initial pairwise constraint matrix Z is used.

[0014] Introduce projection matrix N for each view (v) Using the projection matrix N (v)The shared label constraint matrix A generates the representation matrix V for each view. (v) =AN (v) By using point constraints to label the low-dimensional representation of the samples, we can ensure that the samples are consistent with the true labels.

[0015] The initialization of the correlation matrix represents the linear mapping relationship from the original data to the low-dimensional space, and the mean of each view representation matrix is ​​used to generate the consensus matrix.

[0016] In this scheme, an initial weight matrix is ​​constructed for each view sample data based on nearest neighbor relationships and kernel functions. The final pairwise constraint matrix is ​​then generated through vertical and horizontal constraint propagation algorithms. Specifically:

[0017] For the data matrix of the v-th view, the similarity between samples is calculated using a Gaussian kernel function, and for each sample x... i Obtain a preset number of nearest neighbors based on sample x. i A sparse similarity matrix is ​​constructed based on the relationship between the nearest neighbor samples, and an initial weight matrix is ​​constructed through symmetric normalization.

[0018] Calculate the graph Laplacian matrix based on the initial weight matrix, and propagate the vertical constraint matrix F. v and horizontal constraint propagation matrix F h The elements are set to random values ​​between [-1, 1], and initialized as a propagation matrix for pairwise constraint propagation;

[0019] The vertical constraint propagation matrix F is iteratively updated using the Graph Laplacian matrix and preset propagation parameters. v and horizontal constraint propagation matrix F h According to the vertical constraint propagation matrix F v and horizontal constraint propagation matrix F h After iteration until convergence, the extreme values ​​are obtained, and the final pairwise constraint matrix is ​​generated.

[0020] In this scheme, the weight matrix is ​​updated to enhance the correlation between samples of the same type. The corresponding graph Laplacian matrix and diagonal matrix are calculated as follows:

[0021] Obtain the initial weight matrix and the final pairwise constraint matrix. According to the update rule, obtain the updated weight matrix to enhance the association of similar classes. Use the updated weight matrix to calculate the diagonal matrix, and construct the updated graph Laplacian matrix based on the diagonal matrix and the updated weight matrix.

[0022] The update rule is expressed as follows:

[0023]

[0024] in s represents the elements in the updated weight matrix. ijThis represents the elements in the initial weight matrix. This represents the elements in the final pairwise constraint matrix, if Then increase Strengthening similar associations, if Then decrease Weaken the association between similar categories.

[0025] In this scheme, the projection matrix is ​​fixed during alternating iterations, and the correlation matrix is ​​updated using the shared label constraint matrix. The correlation matrix is ​​also fixed, and the projection matrix is ​​updated using the graph Laplacian matrix and the diagonal matrix. Specifically:

[0026] The objective function is optimized through alternating iterations, with the projection matrix N fixed during the iterations. (v) Update the association matrix W of each view. (v) To minimize reconstruction error and point constraint loss, known category label information is embedded into the association matrix by sharing the label constraint matrix and projection matrix, ensuring the consistency of the low-dimensional representation of known category label samples. The association matrix W for each view... (v) The update formula is expressed as:

[0027]

[0028] in, K represents the elements in the correlation matrix. (v) Let A represent the kernel matrix and A represent the shared label constraint matrix.

[0029] The correlation matrix W is fixed during the iteration. (v) Update the projection matrix N of each view. (v) Geometric information encoded by the Laplacian matrix guides unknown category labeled samples to cluster towards the correct category. The projection matrices N of each view... (v) The update formula is:

[0030]

[0031] in, This represents the elements in the projection matrix, where α and β represent regularization parameters. V represents the updated weight matrix. * Let A represent the updated consensus matrix, and let A represent the shared label constraint matrix.

[0032] In this scheme, after each iteration, the consensus matrix is ​​updated based on the projection matrix of each view, specifically as follows:

[0033] Obtain the updated projection matrix N for each view. (v) and shared label constraint matrix N (v) Calculate the representation matrix for each view, average the representation matrices of all views, fuse multi-view information, eliminate differences between views, and obtain a unified final consensus matrix V.* , is represented as:

[0034]

[0035] Where n v represents the total number of views, and v represents the number of view items.

[0036] In this scheme, the objective function is optimized through alternating iterations, and the objective function is expressed as:

[0037]

[0038] Where X (v) W represents the data matrix of the v-th view. (v) N represents the association matrix of the v-th view. (v) V represents the projection matrix of the v-th view. * Let represent the final consensus matrix for multi-view fusion, where α and β represent regularization parameters. Let W represent the graph Laplacian matrix. (v) N (v) V * ≥0 indicates a non-negativity constraint, ensuring the decomposition results are interpretable. v The total number of views is represented by Tr, which represents the trace of the matrix and is used for scalarization of the manifold regularization term.

[0039] In this scheme, after the iteration is complete, the final consensus matrix is ​​used as a low-dimensional representation of the view. The k-means algorithm is then used for clustering to output the final category classification results, specifically:

[0040] Obtain the final consensus matrix V * and a preset number of clustering categories, for the final consensus matrix V * Normalization is performed, sample points are selected as initial cluster centers according to the preset number of cluster categories, the Euclidean distance between each sample and all initial cluster centers is calculated, and the sample is assigned to the nearest initial cluster center to generate a data cluster.

[0041] For each data cluster, the mean of all samples in the data cluster is recalculated to update the cluster center. The process stops when the change in the cluster center is less than the threshold or when the maximum number of iterations is reached. The final category label of each sample is obtained based on the final clustering result.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention proposes a semi-supervised multi-view clustering model based on dual-constraint concept decomposition. By innovatively introducing a dual-supervision mechanism combining point constraints and pairwise constraints, point constraints directly act on labeled samples to ensure cluster consistency; simultaneously, pairwise constraints construct "must-link" and "cannot-link" relationships between samples, propagating labeling information to unlabeled data, capturing the inherent geometric structure of the data, and expanding the scope of supervision. Furthermore, based on the concept decomposition framework, it is compatible with data containing negative values ​​and kernel methods, enabling effective processing of complex multi-view data. Ultimately, this comprehensively improves the performance of multi-view clustering tasks, overcoming the bottlenecks of existing technologies in clustering accuracy, data adaptability, and the efficiency of supervision information utilization. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0045] Figure 1 A flowchart of a semi-supervised multi-view clustering method based on dual-constraint concept decomposition is shown;

[0046] Figure 2 A framework diagram of a semi-supervised multi-view clustering method based on dual-constraint concept decomposition is shown.

[0047] Figure 3 A block diagram of a semi-supervised multi-view clustering system based on dual-constraint concept decomposition is shown. Detailed Implementation

[0048] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0050] Figure 1 A flowchart of a semi-supervised multi-view clustering method based on dual-constraint concept decomposition is shown.

[0051] like Figure 1 As shown, this embodiment provides a semi-supervised multi-view clustering method based on dual-constraint concept decomposition, including:

[0052] S102, Obtain a multi-view dataset containing multiple view samples, some of which have known labels, and initialize the multi-view shared label constraint matrix, pairwise constraint matrix, representation matrix, association matrix, and consensus matrix in the multi-view dataset;

[0053] S104: For each view, construct an initial weight matrix based on nearest neighbor relationships and kernel functions, generate the final pairwise constraint matrix through vertical and horizontal constraint propagation algorithms, update the weight matrix to enhance the correlation of similar samples, and calculate the corresponding graph Laplacian matrix and diagonal matrix.

[0054] S106, in the alternating iteration, the fixed projection matrix is ​​updated using the shared label constraint matrix, the fixed association matrix is ​​updated using the graph Laplacian matrix and the diagonal matrix, and the consensus matrix is ​​updated based on the projection matrix of each view after each iteration.

[0055] S108. After the iteration is completed, the final consensus matrix is ​​used as a low-dimensional representation of the view. The k-means algorithm is used for clustering to output the final category division result.

[0056] It should be noted that, as Figure 2 As shown, this scheme combines point constraints with pairwise constraints to construct the LDC algorithm (a dual-constraint algorithm based on prior label information) with a complementary supervision mechanism, thereby maximizing the utilization of limited supervision information and progressive propagation. On the other hand, relying on the concept decomposition framework, it breaks through the limitations of traditional methods on data non-negativity and combines kernel methods to handle complex data distributions, making it suitable for multi-view data scenarios such as news classification, fake news detection, and user interest mining.

[0057] Input multi-view dataset X = [X (1) ,X (2) ,…,X (v) ,…,X (nv) ], where X (v) Let v represent the data matrix of the v-th view. The multi-view dataset contains n views, each view has m feature dimensions, the first p samples have known class labels, and the remaining np samples have no labels. The number of classes is c.

[0058] Encoding known category label information C for samples with known category labels p×c Using the identity matrix I n-p This represents the initial free representation of an unknown-labeled sample, encoded based on the known label portions. p×c The unknown label part indicates I n-p Construct a multi-view shared label constraint matrix A ij =1 indicates that the i-th sample belongs to the j-th class, Aij =0 indicates other cases. The shared label constraint matrix uses known label information as point constraints, and unlabeled samples are gradually aligned with the true distribution through optimization.

[0059] Pair constraints are constructed based on samples with known category labels. The constraint relationship is established by determining whether the samples in a pair are of the same class. The initial pair constraint matrix Z provides the initial supervision signal for subsequent pair constraint propagation. Each element is defined as follows:

[0060]

[0061] Where M represents x i With x j If they share the same label, C indicates they belong to different labels, and 0 indicates other cases.

[0062] Introduce projection matrix N for each view (v) Using the projection matrix N (v) The shared label constraint matrix A generates the representation matrix V for each view. (v) =AN (v) By using point constraints to label the low-dimensional representation of the samples, we ensure that the samples are consistent with the true labels; we initialize the association matrix to represent the linear mapping relationship from the original data to the low-dimensional space, and find the following approximate representation X≈XWV in the concept decomposition. T Where X is the original data, W is the association matrix representing the degree of relevance of data points to the concept, and V is the representation matrix. Additionally, the average of the representation matrices of each view is used to generate a consensus matrix, and all view information is integrated to obtain a unified low-dimensional representation.

[0063] It should be noted that, assuming Propagate pairwise constraint matrices, with absolute values ​​in |f ij | ≤ 1. When f ij >0, indicating (x i ,x j ) is a mandatory link constraint. ij <0 means the constraint cannot be linked, while |f ij | represents (x) i ,x j () is the confidence score for constraints that must be linked or cannot be linked. This is determined by defining its weight matrix. To compute the p-nearest neighbor graph

[0064]

[0065] For each data point x i Find its p-nearest neighbor (excluding itself). If x i For x jA nearest neighbor is used to calculate the similarity between samples for the data matrix of the v-th view using a Gaussian kernel function, with weights s. ij a(x) i ,x j )=exp(-‖x i -x j || 2 / t)(where t is the kernel bandwidth parameter); otherwise, let s ij =0. Additionally, to make the weight matrix symmetric, S = (S + S) T ) / 2;

[0066] After obtaining the symmetric weight matrix, calculate the graph Laplacian matrix. The matrix D involved is a diagonal matrix. Its construction rule is that the i-th diagonal element of D is obtained by considering all elements s of all j. ij It is determined by summation, denoted as ∑ j s ij .

[0067] The vertical constraint propagation matrix F v and horizontal constraint propagation matrix F h The elements are set to random values ​​between [-1, 1], and the vertical constraint propagation matrix F is iteratively updated using the graph Laplacian matrix and preset propagation parameters. v and horizontal constraint propagation matrix F h According to the vertical constraint propagation matrix F v and horizontal constraint propagation matrix F h After iteration until convergence, the extreme values ​​are obtained, and the final pairwise constraint matrix is ​​generated.

[0068] F v (t+1)=αLF v (t)+(1-α)Z

[0069]

[0070] in, It is {F v The limit of (t)} It is {F h The limit of F (t)}. * The final configuration of the paired constraints established during propagation reflects their integrated representation. It is worth noting that F... * This represents a set of confidence scores that combine a pair of constraints and their associations.

[0071] Obtain the initial weight matrix and the final pairwise constraint matrix. Based on the update rules, obtain the updated weight matrix to enhance class-specific associations. Then, use the updated weight matrix to calculate the diagonal matrix. And an updated graph Laplacian matrix is ​​constructed based on the diagonal matrix and the updated weight matrix.

[0072] The update rule is expressed as follows:

[0073]

[0074] in s represents the elements in the updated weight matrix. ij This represents the elements in the initial weight matrix. This represents the elements in the final pairwise constraint matrix, if Then increase Strengthening similar associations, if Then decrease Weaken the association between similar categories.

[0075] It should be noted that the objective function is optimized through alternating iterations, with the projection matrix N fixed during the iterations. (v) Update the association matrix W of each view. (v) To minimize reconstruction error and point constraint loss, known category label information is embedded into the association matrix by sharing the label constraint matrix and projection matrix, ensuring the consistency of the low-dimensional representation of known category label samples. The association matrix W for each view... (v) The update formula is expressed as:

[0076]

[0077] in, K represents the elements in the correlation matrix. (v) Let A represent the kernel matrix and A represent the shared label constraint matrix.

[0078] The correlation matrix W is fixed during the iteration. (v) Update the projection matrix N of each view. (v) Geometric information encoded by the Laplacian matrix guides unknown category labeled samples to cluster towards the correct category. The projection matrices N of each view... (v) The update formula is:

[0079]

[0080] in, This represents the elements in the projection matrix, where α and β represent regularization parameters. V represents the updated weight matrix. * Let A represent the updated consensus matrix, and let A represent the shared label constraint matrix.

[0081] Obtain the updated projection matrix N for each view. (v) and shared label constraint matrix N (v)Calculate the representation matrix for each view, average the representation matrices of all views, fuse multi-view information, eliminate differences between views, and obtain a unified final consensus matrix V. * , is represented as:

[0082]

[0083] Where n v represents the total number of views, and v represents the number of view items.

[0084] It should be noted that by iteratively optimizing the objective function, and by minimizing the reconstruction error, maintaining geometric consistency (manifold regularization), and multi-view collaborative optimization (consensus constraint), a discriminative low-dimensional representation is learned. The objective function is expressed as:

[0085]

[0086] Where X (v) W represents the data matrix of the v-th view. (v) N represents the association matrix of the v-th view. (v) V represents the projection matrix of the v-th view. * Let represent the final consensus matrix for multi-view fusion, where α and β represent regularization parameters. Let W represent the graph Laplacian matrix. (v) N (v) V * ≥0 indicates a non-negativity constraint, ensuring the decomposition results are interpretable. v The total number of views is represented by Tr, which represents the trace of the matrix and is used for scalarization of the manifold regularization term.

[0087] By introducing a label constraint matrix A and an auxiliary matrix N shared across multiple views (v) The concept decomposition is obtained by applying point constraints as hard constraints under multiple views, and the reconstruction error is measured using the Frobenius norm. Ensure low-dimensional representation of AN (v) It can effectively reconstruct the original data X (v) Pairwise constraint information is reflected through modifications to the weight matrix; data points within the same class have relatively high correlation values, while data points from different classes have lower correlation values. The prior label-based dual-constraint algorithm combines point constraints and pairwise constraint propagation, fully utilizing limited prior label information and unifying it within the concept decomposition framework. The manifold regularization term is... Using the Laplace matrix Constrain similar samples to be close in distance in low-dimensional space. (Multi-view) Figure 1 Coherence terms The consistency error is measured using the Frobenius norm, forcing a low-dimensional representation AN for each view.(v) Approximating the final consensus matrix. Point constraints ensure consistency among known labeled samples, while pairwise constraints propagate supervisory information to unlabeled samples. Combining multi-view consensus matrices with complementary features effectively improves clustering accuracy.

[0088] It should be noted that obtaining the final consensus matrix V * and a preset number of clustering categories, for the final consensus matrix V * Normalization is performed, and sample points are selected as initial cluster centers according to the preset number of cluster categories. The Euclidean distance between each sample and all initial cluster centers is calculated, and the sample is assigned to the nearest initial cluster center to generate a data cluster. For each data cluster, the mean of all samples in the data cluster is recalculated to update the cluster center. The process stops when the change in cluster center is less than the threshold or the maximum number of iterations is reached. The final category label of each sample is obtained based on the final clustering result.

[0089] In a preferred embodiment of this application, the BBC news dataset includes four views (word frequency, TF-IDF, topic model features, and named entity statistics). The sample comprises 685 news articles, categorized into five types (politics, business, technology, entertainment, and sports). Approximately 68 articles (10% of the samples) are randomly selected to provide ground truth labels; the remainder are unlabeled. The first 68 rows correspond to known labels (e.g., political category sample A). ij =1, the rest A ij =0), the last 617 rows are an identity matrix. From the known labels, 50 pairs of "must-link" (similar samples) and 50 pairs of "cannot-link" (different samples) are randomly generated. Initial similarity is calculated based on the Gaussian kernel function, with a nearest neighbor count p=5, and then generated... Strengthen the association between similar samples.

[0090] Regularization coefficients α = 10, β = 100, and latent space dimension k = 5 (consistent with the number of classes). N is fixed. (v) Update W (v) Minimize reconstruction error, fix W (v) ,use Update N (v) This disseminates supervisory information to unlabeled samples and updates the consensus matrix. The objective function changes by less than 10 -6 Alternatively, the iteration can be terminated after 100 iterations. Then, the consensus matrix is ​​updated for clustering, 5 cluster centers are initialized, and the iteration continues until convergence, outputting the category labels for 685 news articles.

[0091] Figure 3 A block diagram of a semi-supervised multi-view clustering system based on dual-constraint concept decomposition is shown.

[0092] The second embodiment of the present invention provides a semi-supervised multi-view clustering system 3 based on dual-constraint concept decomposition, including: a memory 31 and a processor 32. The memory includes a program for a semi-supervised multi-view clustering method based on dual-constraint concept decomposition. When the program for the semi-supervised multi-view clustering method based on dual-constraint concept decomposition is executed by the processor, the steps of the semi-supervised multi-view clustering method based on dual-constraint concept decomposition are implemented.

[0093] A third aspect of the present invention provides a computer-readable storage medium comprising a program for a semi-supervised multi-view clustering method based on dual-constraint concept decomposition, wherein when the program for the semi-supervised multi-view clustering method based on dual-constraint concept decomposition is executed by a processor, the program implements the steps of the semi-supervised multi-view clustering method based on dual-constraint concept decomposition.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms. Furthermore, in the various embodiments of the present invention, all functional units can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0095] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A semi-supervised multi-view clustering method based on dual-constraint concept decomposition, characterized in that, Includes the following steps: Obtain a multi-view dataset containing multiple view samples, some of which have known labels. Initialize the multi-view shared label constraint matrix, pairwise constraint matrix, representation matrix, association matrix, and consensus matrix in the multi-view dataset. For each view, an initial weight matrix is ​​constructed based on nearest neighbor relationships and kernel functions. The final pairwise constraint matrix is ​​generated through vertical and horizontal constraint propagation algorithms. The weight matrix is ​​updated to enhance the correlation of samples of the same type. The corresponding graph Laplacian matrix and diagonal matrix are calculated. In the alternating iteration, the projection matrix is ​​fixed and the association matrix is ​​updated using the shared label constraint matrix. The association matrix is ​​fixed and the projection matrix is ​​updated using the graph Laplacian matrix and the diagonal matrix. After each iteration, the consensus matrix is ​​updated based on the projection matrix of each view. After the iteration is completed, the final consensus matrix is ​​used as a low-dimensional representation of the view, and the k-means algorithm is used for clustering to output the final category division results; Obtain a multi-view dataset containing sample data from multiple views, some of which have known labels. Initialize the shared label constraint matrix, representation matrix, projection matrix, and consensus matrix of the view sample data in the multi-view dataset as follows: Input multi-view dataset ,in Indicates the first The data matrix contains n views, each with features including text term frequency features, TF-IDF features, topic model features, and named entity statistics. Feature dimension, front One sample has a known class label, the rest There are no labeled samples, and the number of categories is [number missing]. ; Encoding known category label information for samples with known category labels Using the identity matrix This represents the initial free representation of an unknown-label sample, encoded based on the known labels. and the unknown label part indicates Construct a multi-view shared label constraint matrix ; Based on samples with known category labels, pairwise constraints are constructed. The constraint relationship is established by determining whether samples in a pair belong to the same class. An initial pairwise constraint matrix is ​​then generated. ; Introduce a projection matrix for each view Using the projection matrix and the shared label constraint matrix Generate a representation matrix for each view. By using point constraints to label the low-dimensional representation of the samples, we can ensure that the samples are consistent with the true labels. The initialization of the correlation matrix represents the linear mapping relationship from the original data to the low-dimensional space, and the mean of each view representation matrix is ​​used to generate the consensus matrix.

2. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 1, characterized in that, For each view sample data, an initial weight matrix is ​​constructed based on nearest neighbor relationships and kernel functions. The final pairwise constraint matrix is ​​then generated using vertical and horizontal constraint propagation algorithms, specifically: For the The data matrix of each view is used to calculate the similarity between samples using a Gaussian kernel function. Obtain a preset number of nearest neighbors based on samples. A sparse similarity matrix is ​​constructed based on the relationship between the nearest neighbor samples, and an initial weight matrix is ​​constructed through symmetric normalization. Calculate the graph Laplacian matrix based on the initial weight matrix, and propagate the vertical constraint matrix. and horizontal constraint propagation matrix The elements are set to random values ​​between [-1, 1], and initialized as a propagation matrix for pairwise constraint propagation; The vertical constraint propagation matrix is ​​iteratively updated using the Graph Laplacian matrix and preset propagation parameters. and horizontal constraint propagation matrix According to the vertical constraint propagation matrix and horizontal constraint propagation matrix After iteration until convergence, the extreme values ​​are obtained, and the final pairwise constraint matrix is ​​generated.

3. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 2, characterized in that, To enhance the correlation between similar samples, the weight matrix is ​​updated. The corresponding graph Laplacian matrix and diagonal matrix are calculated as follows: Obtain the initial weight matrix and the final pairwise constraint matrix. According to the update rule, obtain the updated weight matrix to enhance the association of similar classes. Use the updated weight matrix to calculate the diagonal matrix, and construct the updated graph Laplacian matrix based on the diagonal matrix and the updated weight matrix. The update rule is expressed as follows: ， in This represents the elements in the updated weight matrix. This represents the elements in the initial weight matrix. This represents the elements in the final pairwise constraint matrix, if Then increase Strengthening similar associations, if Then decrease Weaken the association between similar categories.

4. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 1, characterized in that, In the alternating iterations, the projection matrix is ​​fixed, and the incidence matrix is ​​updated using the shared label constraint matrix. The incidence matrix is ​​also fixed, and the projection matrix is ​​updated using the graph Laplacian matrix and the diagonal matrix. Specifically: The objective function is optimized through alternating iterations, with the projection matrix fixed during the iterations. Update the association matrix of each view. To minimize reconstruction error and point constraint loss, known category label information is embedded into the association matrix by sharing the label constraint matrix and projection matrix, ensuring the consistency of the low-dimensional representation of known category label samples. The association matrices of each view... The update formula is expressed as: ， in, Represents the elements in the correlation matrix. Represents the kernel matrix. Represents the shared label constraint matrix; Fix the incidence matrix during iteration Update the projection matrices of each view. By using geometric information encoded by the Laplacian matrix, unknown category labeled samples are guided to cluster towards the correct category, and the projection matrices of each view are used. The update formula is: ， in, Represents the elements in the projection matrix. Represents the regularization parameter. This represents the updated weight matrix. This indicates updating the consensus matrix. This represents the shared label constraint matrix.

5. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 4, characterized in that, After each iteration, the consensus matrix is ​​updated based on the projection matrix of each view, specifically as follows: Get the updated projection matrix for each view and shared label constraint matrix The representation matrix of each view is calculated, the average of the representation matrices of all views is taken, the information from multiple views is fused, the differences between views are eliminated, and a unified final consensus matrix is ​​obtained. , is represented as: ， in Indicates the total number of views. Indicates the number of view items.

6. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 4, characterized in that, The objective function is optimized through alternating iterations, and the objective function is expressed as: ， in Indicates the first Data matrix of views, Indicates the first The association matrix of each view, Indicates the first Projection matrix of each view, This represents the final consensus matrix for multi-view fusion. , Represents the regularization parameter. Represents the graph Laplace matrix. This indicates a nonnegativity constraint, ensuring that the decomposition results are interpretable. Indicates the total number of views. Represents the trace of a matrix, used for scalarization of manifold regularization terms.

7. The semi-supervised multi-view clustering method based on dual-constraint concept decomposition according to claim 1, characterized in that, After iteration, the final consensus matrix is ​​used as a low-dimensional representation of the view. The k-means algorithm is then used for clustering to output the final category classification results, specifically: Obtain the final consensus matrix and a preset number of clustering categories, for the final consensus matrix Normalization is performed, sample points are selected as initial cluster centers according to the preset number of cluster categories, the Euclidean distance between each sample and all initial cluster centers is calculated, and the sample is assigned to the nearest initial cluster center to generate a data cluster. For each data cluster, the mean of all samples in the cluster is recalculated to update the cluster center. The process stops when the change in the cluster center is less than the threshold or when the maximum number of iterations is reached. The final category label of each sample is obtained based on the final clustering result and used for the classification processing of text data.

Images