Method and related apparatus for consumer classification based on multi-view clustering

Abstract

The application discloses a multi-view clustering-based consumer classification method and related device, the method comprising: performing feature extraction on a multi-view consumer data set of a consumer to obtain a multi-view data matrix; analyzing the reverse neighbor relationship between users based on a K reverse neighbor algorithm according to the multi-view data matrix to obtain a K reverse neighbor data set; calculating the Gaussian kernel similarity between any two user consumption operations in the K reverse neighbor data set to construct a metric matrix view; performing similarity fusion analysis of the users based on a preset diagonal line Mahalanobis matrix according to the metric matrix view to obtain a fused similarity matrix graph; and performing user clustering analysis on the fused similarity matrix graph through Laplacian rank constraint to obtain a consumer clustering result. Therefore, the application can solve the technical problems that the prior art cannot exclude noise interference, the analysis of consumer behavior lacks pertinence, and the separation of graph analysis and clustering operation leads to a lack of accuracy of the result.

Description

Technical Field

[0001] This application relates to the field of cluster analysis technology, and in particular to a consumer user classification method and related apparatus based on multi-view clustering. Background Technology

[0002] With the widespread adoption of internet technology and social media, consumer behavior and preferences have become increasingly complex and diverse. Consumer behavior analysis, as a crucial component of marketing, has become key to businesses' successful development and implementation of marketing strategies. In general consumption scenarios, customer behavior data is transformed into feature data. Cluster analysis is then used to categorize target customers into groups, identifying valuable specific groups and uncovering potential customers. This allows for product recommendations based on the consumption preferences of each customer or group, increasing sales volume and promoting consumer spending.

[0003] Clustering is a paradigm that divides samples into different subgroups based on the similarity between them. It is a commonly used method in machine learning, pattern recognition, and data mining. Once subgroups are obtained through clustering, many subsequent analytical tasks can be performed to achieve different final goals. Multi-view clustering, on the other hand, involves grouping multiple sets of features under each topic that need to be studied and analyzed separately, and then aggregating these views to achieve the purpose of grouping.

[0004] Existing multi-view clustering analysis techniques for consumer classification, whether based on fully connected complete graphs or K-nearest neighbor thresholds, cannot eliminate noise and outliers caused by consumer mis-triggers, resulting in inaccurate matrices or graphs. Furthermore, they treat consumer clicks, purchases, favorites, and add-to-cart behaviors equally, ignoring the impact of different consumer behaviors on consumer classification. In addition, separating the construction of similarity graphs from the clustering results also leads to inaccurate final results. Summary of the Invention

[0005] This application provides a consumer user classification method and related apparatus based on multi-view clustering, which solves the technical problems that existing technologies cannot eliminate noise interference, lack specificity in the analysis of consumer behavior, and result in inaccurate results due to separation graph analysis and clustering operations.

[0006] In view of this, the first aspect of this application provides a consumer user classification method based on multi-view clustering, including:

[0007] Feature extraction is performed on the multi-view consumption dataset of consumers to obtain a multi-view data matrix;

[0008] Based on the K-reverse nearest neighbor algorithm, the reverse nearest neighbor relationship between users is analyzed according to the multi-view data matrix to obtain the K-reverse nearest neighbor dataset;

[0009] Calculate the Gaussian kernel similarity between any two user consumption operations in the K-reverse nearest neighbor dataset, and construct a metric matrix view;

[0010] Based on the preset diagonal Mahalanobis matrix, the similarity fusion analysis of users is performed according to the metric matrix view to obtain a fused similarity matrix diagram;

[0011] User clustering analysis was performed on the fused similarity matrix graph using Laplace rank constraints to obtain consumer user clustering results.

[0012] Preferably, the K-reverse nearest neighbor algorithm analyzes the reverse nearest neighbor relationships between users based on the multi-view data matrix to obtain a K-reverse nearest neighbor dataset, including:

[0013] Based on the K-reverse nearest neighbor algorithm, the K-reverse nearest neighbor dataset is obtained by finding the reverse nearest neighbor samples of each user's consumption operation sample that belong to the nearest neighbor of other users' consumption operation samples in each consumption behavior according to the multi-view data matrix.

[0014] Preferably, the step of performing user similarity fusion analysis based on a preset diagonal Mahalanobis matrix and the metric matrix view to obtain a fused similarity matrix diagram includes:

[0015] Calculate the similarity matrix among users in the multi-view data matrix based on the initial Mahalanobis identity matrix;

[0016] A similarity fusion analysis is performed based on the similarity Mahalanobis matrix, the metric matrix view, and the preset penalty term to obtain an initial fusion matrix diagram.

[0017] A positive semi-definite Markov matrix update operation is performed based on the initial fusion matrix diagram to obtain a preset diagonal Markov matrix.

[0018] Replace the initial Mahalanobis identity matrix with the preset diagonal Mahalanobis matrix, and return to the step of calculating the similarity Mahalanobis matrix between users in the multi-view data matrix based on the initial Mahalanobis identity matrix, until the iteration stop condition is reached, and obtain the fused similarity matrix graph.

[0019] Preferably, the step of performing similarity fusion analysis based on the similarity Mahalanobis matrix, the metric matrix view, and a preset penalty term to obtain an initial fusion matrix diagram includes:

[0020] Based on the similarity Mahalanobis matrix, the metric matrix view, and the preset penalty term, a fusion analysis of similarity is performed to obtain local structure fusion parameters;

[0021] Based on the local structure fusion parameters, a similarity graph corresponding to the metric matrix view is constructed to obtain the initial fusion matrix graph.

[0022] The second aspect of this application provides a consumer user classification device based on multi-view clustering, including:

[0023] The feature extraction unit is used to extract features from the multi-view consumer dataset to obtain a multi-view data matrix;

[0024] The nearest neighbor analysis unit is used to analyze the reverse nearest neighbor relationship between users based on the K reverse nearest neighbor algorithm according to the multi-view data matrix, and obtain the K reverse nearest neighbor dataset.

[0025] The similarity calculation unit is used to calculate the Gaussian kernel similarity between any two user consumption operations in the K reverse nearest neighbor dataset and construct a metric matrix view;

[0026] The similarity fusion unit is used to perform user similarity fusion analysis based on the preset diagonal Mahalanobis matrix and the metric matrix view to obtain a fused similarity matrix diagram;

[0027] The clustering analysis unit is used to perform user clustering analysis on the fused similarity matrix graph through Laplace rank constraints to obtain consumer user clustering results.

[0028] Preferably, the nearest neighbor analysis unit is specifically used for:

[0029] Based on the K-reverse nearest neighbor algorithm, the K-reverse nearest neighbor dataset is obtained by finding the reverse nearest neighbor samples of each user's consumption operation sample that belong to the nearest neighbor of other users' consumption operation samples in each consumption behavior according to the multi-view data matrix.

[0030] Preferably, the similarity fusion unit includes:

[0031] A similarity calculation subunit is used to calculate the similarity Mahalanobis matrix between users in the multi-view data matrix based on the initial Mahalanobis identity matrix.

[0032] The fusion analysis subunit is used to perform similarity fusion analysis based on the similarity Mahalanobis matrix, the metric matrix view, and a preset penalty term to obtain an initial fusion matrix diagram.

[0033] The matrix update subunit is used to perform a positive semidefinite Markov matrix update operation based on the initial fusion matrix diagram to obtain a preset diagonal Markov matrix.

[0034] The iterative calculation subunit is used to replace the initial Mahalanobis identity matrix with the preset diagonal Mahalanobis matrix and trigger the similarity calculation subunit until the iteration stop condition is reached to obtain the fused similarity matrix graph.

[0035] Preferably, the fusion analysis subunit is specifically used for:

[0036] Based on the similarity Mahalanobis matrix, the metric matrix view, and the preset penalty term, a fusion analysis of similarity is performed to obtain local structure fusion parameters;

[0037] Based on the local structure fusion parameters, a similarity graph corresponding to the metric matrix view is constructed to obtain the initial fusion matrix graph.

[0038] A third aspect of this application provides a consumer user classification device based on multi-view clustering, the device including a processor and a memory;

[0039] The memory is used to store program code and transmit the program code to the processor;

[0040] The processor is configured to execute the consumer user classification method based on multi-view clustering as described in the first aspect, according to the instructions in the program code.

[0041] The fourth aspect of this application provides a computer-readable storage medium for storing program code for executing the consumer user classification method based on multi-view clustering described in the first aspect.

[0042] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0043] This application provides a consumer user classification method based on multi-view clustering, including: extracting features from a multi-view consumer dataset to obtain a multi-view data matrix; analyzing the reverse nearest neighbor relationships between users based on the K-reverse nearest neighbor algorithm according to the multi-view data matrix to obtain a K-reverse nearest neighbor dataset; calculating the Gaussian kernel similarity between any two user consumption operations in the K-reverse nearest neighbor dataset to construct a metric matrix view; performing user similarity fusion analysis based on a preset diagonal Mahalanobis matrix according to the metric matrix view to obtain a fused similarity matrix graph; and performing user clustering analysis on the fused similarity matrix graph through Laplace rank constraints to obtain consumer user clustering results.

[0044] The consumer user classification method based on multi-view clustering provided in this application employs the K-reverse nearest neighbor algorithm to analyze the reverse nearest neighbor relationships between users, which can eliminate the interference of noise and outliers. Furthermore, by calculating the metric matrix and performing similarity fusion analysis based on the diagonal Mahalanobis matrix, different weights can be assigned to different consumption behaviors, ensuring that different consumption behaviors have different impacts on consumer segmentation, which is more in line with reality and can ensure more accurate classification results. In addition, clustering analysis based on Laplace rank constraints combines graph analysis with clustering operations, improving the reliability of classification results. Therefore, this application can solve the technical problems of existing technologies failing to eliminate noise interference, lacking specificity in the analysis of consumption behavior, and lacking accuracy in results due to separating graph analysis and clustering operations. Attached Figure Description

[0045] Figure 1 A flowchart illustrating the consumer user classification method based on multi-view clustering provided in this application embodiment;

[0046] Figure 2 A schematic diagram of the structure of a consumer user classification device based on multi-view clustering provided in an embodiment of this application;

[0047] Figure 3 A schematic diagram of the nearest neighbor relationship of sample points based on the K-nearest neighbor algorithm provided in the embodiments of this application;

[0048] Figure 4 The similarity matrix diagram provided in this application represents the similarity association between sample points. Detailed Implementation

[0049] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0050] For easier understanding, please refer to Figure 1 The embodiments of the consumer user classification method based on multi-view clustering provided in this application include:

[0051] Step 101: Extract features from the multi-view consumer dataset to obtain a multi-view data matrix.

[0052] It should be noted that the dataset formed by the multi-view data matrix can be expressed as X. v ∈d v ×n, and Where n represents the user ID, v represents the consumption behavior view, and d represents the quantity of goods. Let be a d-dimensional vector representing the set of operations performed by the nth consumer on all products within the v-th consumer behavior view. This embodiment aims to classify consumers based on their consumption behavior, grouping similar samples into one category and dispersing dissimilar samples.

[0053] Step 102: Based on the K-reverse nearest neighbor algorithm, analyze the reverse nearest neighbor relationships between users according to the multi-view data matrix to obtain the K-reverse nearest neighbor dataset.

[0054] Further, step 102 includes:

[0055] Based on the K-reverse nearest neighbor algorithm, the K-reverse nearest neighbor dataset is obtained by finding the reverse nearest neighbor samples of each user's consumption operation sample that belong to the nearest neighbor of other users' consumption operation samples in each consumption behavior according to the multi-view data matrix.

[0056] It's important to note that the difference between the K-reverse nearest neighbor algorithm and the K-nearest neighbor algorithm is that K-nearest neighbor starts from a sample point and finds which data points are its neighbors; while K-reverse nearest neighbor analyzes which other sample points are the neighbors of a given sample point. For any given dataset D, including x... p Let represent the set vector of operations performed by the p-th consumer on all products in any consumer behavior view. This set can be expressed using the following formula:

[0057] RNN K (x p ,D)={x q |x q ∈D\{x q},x p ∈NN K (x q ,D)}

[0058] Among them, NN K (x q (,D) represents x in dataset D q The K-nearest neighbor dataset is obtained by using this method to calculate the K-nearest neighbor sample set for all users in each consumption behavior. Furthermore, the K-nearest neighbor sample set may be greater than K, equal to K, or less than K, and may also be empty. If the K-nearest neighbor sample set is empty, it means that the sample point is an isolated point in the distribution and does not belong to the nearest neighbors of any other sample point. For an example, please refer to [link to example]. Figure 3 The points in the circle represent the K-nearest neighbor set, where NN K (a) = {b, d, e}, NN K (b) = {c, d, e}, NN K (c) = {b, d, e}, NN K (d) = {a, b, e}, NN K (e) = {a, b, d}, NN K (f) = {a, b, d, e}; the corresponding K-reverse nearest neighbor sample set is represented by an RNN. K (a) = {d, e, f}, RNN K (b) = {a,c,d,e,f}, RNN K (c) = {b}, RNN K (d) = {a, b, c, e, f}, RNN K(e) = {a, b, c, d, f}, NN K (f)={Φ}; It can be found that the sample point f is too far away from other sample points, so it is an outlier point, and its K-nearest neighbor sample set is an empty set.

[0059] Step 103: Calculate the Gaussian kernel similarity between any two user consumption operations in the K-reverse nearest neighbor dataset and construct the metric matrix view.

[0060] It should be noted that this embodiment uses a Gaussian kernel to calculate the similarity between two users' operations on the same product. Overall, it analyzes the similarity between the consumption operations of two consumers on the product. The operation vectors of any two users in any consumption behavior are expressed as follows: Similarity calculation is expressed as:

[0061]

[0062] Where d is the number of goods, d α Let σ represent the total number of goods, and σ be the variance of the Gaussian kernel function. For two users' actions on the same item, the differences are calculated and then summed to obtain the consumption operation results for all items. Based on this, the similarity between the two users' consumption operations is obtained. Similarity can be calculated for all users in this way, thus constructing a metric matrix, i.e., the metric matrix view S.

[0063] It is understandable that the greater the similarity between two user samples, the closer they are, and vice versa. To standardize the similarity range, facilitate comparative analysis, and reduce computational load, this embodiment can also normalize the obtained similarity scores so that the sum of the sample similarities is 1. To avoid trivial solutions, i.e., a sample having a similarity of 1 to itself, resulting in a similarity of 0 to other samples, this embodiment directly defines S... ii =0.

[0064] Step 104: Based on the preset diagonal Mahalanobis matrix, perform user similarity fusion analysis according to the metric matrix view to obtain the fused similarity matrix diagram.

[0065] Further, step 104 includes:

[0066] Calculate the similarity matrix among users in the multi-view data matrix based on the initial Mahalanobis identity matrix;

[0067] A similarity fusion analysis is performed based on the similarity Mahalanobis matrix, the metric matrix view, and the preset penalty term to obtain the initial fusion matrix diagram.

[0068] Based on the initial fusion matrix diagram, a positive semidefinite Markov matrix update operation is performed to obtain a preset diagonal Markov matrix;

[0069] Replace the initial Mahalanobis identity matrix with the preset diagonal Mahalanobis matrix, and return to the steps of calculating the similarity Mahalanobis matrix between users in the multi-view data matrix based on the initial Mahalanobis identity matrix, until the iteration stops and the fused similarity matrix graph is obtained.

[0070] Furthermore, a similarity fusion analysis is performed based on the similarity Mahalanobis matrix, the metric matrix view, and a preset penalty term to obtain an initial fusion matrix diagram, including:

[0071] Based on the similarity matrix, metric matrix view and preset penalty term, a fusion analysis of similarity is performed to obtain local structure fusion parameters;

[0072] Based on the local structure fusion parameters, a similar graph corresponding to the metric matrix view is constructed to obtain the initial fusion matrix graph.

[0073] It should be noted that there are two reasons for considering using Mahalanobis matrix as a metric to learn similar matrices in this embodiment. First, Mahalanobis matrix is ​​designed to learn matrices that can parameterize the squared distance, and learning Mahalanobis matrix is ​​equivalent to rescaling the learning data. Second, Mahalanobis matrix can be decomposed into matrix product form, making the model easier to solve.

[0074] The initial Mahalanobis identity matrix and the preset diagonal Mahalanobis matrix are positive semi-definite Mahalanobis matrices of the same nature, denoted as M. The initial Mahalanobis identity matrix is ​​the identity matrix when M = I. Through continuous iterative updates, this Mahalanobis matrix will become a diagonal matrix, thereby allowing different weights to be assigned to different axes during the metric learning process, that is, to give different emphasis to different consumption operations of consumers.

[0075] The calculation process for similar Markov matrices is expressed as follows:

[0076]

[0077] Where, x i x j Let |x| represent the consumption operation datasets of consumer i and consumer j in any consumption behavior, respectively. i -x j || M This is the expression form of a similar Markov matrix.

[0078] The process of similarity fusion analysis combining the similarity Mahalanobis matrix, metric matrix view, and preset penalty term can be expressed as follows:

[0079]

[0080]

[0081] in, Let n and m represent the consumption operation datasets of consuming users i and j in consumption behavior v, respectively. n is the total number of user IDs, i.e., the number of users, and m is the total number of consumption behaviors. For example, m = 4 means there are 4 types of consumption behaviors. Let represent the similarity between the i-th user and the j-th user, and the similarity vectors formed by the i-th user and all other users (including itself), respectively. S is the sum of the similarities between user i and all other users (including itself). v This represents the set of metric matrix views corresponding to all consumer behaviors, with one metric matrix view generated for each consumer behavior; β is the penalty coefficient, and U is the initial fusion matrix graph during the iteration process, or the fusion similarity matrix graph after the iteration. The first part of the formula is pairwise similarity learning, used to capture local structural information; the second part is the penalty term.

[0082] Considering that it is a fusion of similar users among multiple consumption behaviors m, the above formula can be expressed as:

[0083]

[0084]

[0085] Among them, u ij u i Let M represent the similarity between the i-th user and the j-th user in the fused similarity matrix graph, and the similarity vector formed by the i-th user and all other users (including itself). vv This represents an updatable or currently being updated diagonal Mahalanobis matrix. The specific update method is expressed as follows:

[0086]

[0087] The updated Mahalanobis matrix is ​​the preset diagonal Mahalanobis matrix. The different elements on the diagonal of this matrix are the weights assigned to the views corresponding to different consumption behaviors. By performing mean weighted calculation on all metric matrix views, the fusion similarity matrix graph U, which reflects the similarity between samples, can be obtained.

[0088] Please see Figure 4 The elements in the fused similarity matrix U are used to describe the similarity values ​​between two samples, that is, the similarity between two consumers. Figure 4 Among the five sample points, the similarity between sample point 1 and sample point 2 is 0.4, the similarity between sample point 2 and sample point 3 is 0.5, the similarity between sample point 3 and sample point 1 is 0.3, sample points 4 and 5 are not related to the previous three sample points, and the similarity between sample points 4 and 5 is 0.6.

[0089] Step 105: Perform user clustering analysis on the fusion similarity matrix graph using Laplace rank constraints to obtain consumer user clustering results.

[0090] by Figure 4 For example, the corresponding adjacency matrix W and degree matrix H can be calculated sequentially; then, based on the adjacency matrix W and degree matrix H, the Laplace matrix L can be obtained. The Laplace matrix L obtained in the example is expressed as:

[0091]

[0092] It can be observed that the data in the Laplace matrix L is divided into two clusters. Based on this property, the fusion similarity matrix graph U can be directly processed based on the Laplace rank constraint to obtain the clustering results.

[0093] Therefore, the similarity fusion analysis formula in this embodiment can be further expressed as:

[0094]

[0095]

[0096] Where c represents the number of clusters, i.e., the number of consumer categories, and L U It depends on the Laplacian matrix of the fused similarity matrix graph U, and the rank constraint is non-linear. The above equation can be further transformed using Ky Fan's theorem:

[0097]

[0098]

[0099] Where γ is the penalty factor, Tr(·) is the trace of the matrix, representing the summation of the elements on the main diagonal of the matrix, and F is the Laplace matrix L. U The embedding matrix obtained by performing singular value decomposition, where I is the identity matrix.

[0100] The consumer user classification method based on multi-view clustering provided in this application employs the K-reverse nearest neighbor algorithm to analyze the reverse nearest neighbor relationships between users, which can eliminate the interference of noise and outliers. Furthermore, by calculating the metric matrix and performing similarity fusion analysis based on the diagonal Mahalanobis matrix, different weights can be assigned to different consumption behaviors, ensuring that different consumption behaviors have different impacts on consumer segmentation, which is more in line with reality and ensures more accurate classification results. In addition, clustering analysis based on Laplace rank constraints combines graph analysis with clustering operations, improving the reliability of classification results. Therefore, this application's embodiments can solve the technical problems of existing technologies failing to eliminate noise interference, lacking specificity in consumer behavior analysis, and resulting in inaccurate results due to separating graph analysis and clustering operations.

[0101] For easier understanding, please refer to Figure 2 This application provides an embodiment of a consumer user classification device based on multi-view clustering, including:

[0102] Feature extraction unit 201 is used to extract features from the multi-view consumer dataset to obtain a multi-view data matrix.

[0103] The nearest neighbor analysis unit 202 is used to analyze the reverse nearest neighbor relationship between users based on the K reverse nearest neighbor algorithm according to the multi-view data matrix, and obtain the K reverse nearest neighbor dataset.

[0104] Similarity calculation unit 203 is used to calculate the Gaussian kernel similarity between any two user consumption operations in the K reverse nearest neighbor dataset and construct a metric matrix view;

[0105] The similarity fusion unit 204 is used to perform user similarity fusion analysis based on a preset diagonal Mahalanobis matrix and a metric matrix view to obtain a fused similarity matrix diagram.

[0106] Clustering analysis unit 205 is used to perform user clustering analysis on the fusion similarity matrix graph through Laplace rank constraints to obtain consumer user clustering results.

[0107] Furthermore, the nearest neighbor analysis unit 202 is specifically used for:

[0108] Based on the K-reverse nearest neighbor algorithm, the K-reverse nearest neighbor dataset is obtained by finding the reverse nearest neighbor samples of each user's consumption operation sample that belong to the nearest neighbor of other users' consumption operation samples in each consumption behavior according to the multi-view data matrix.

[0109] Furthermore, the similarity fusion unit 204 includes:

[0110] Similarity calculation subunit 2041 is used to calculate the similarity Mahalanobis matrix between users in the multi-view data matrix based on the initial Mahalanobis identity matrix;

[0111] The fusion analysis subunit 2042 is used to perform similarity fusion analysis based on the similarity Mahalanobis matrix, the metric matrix view and the preset penalty term to obtain the initial fusion matrix diagram.

[0112] The matrix update subunit 2043 is used to perform a positive semidefinite Markov matrix update operation based on the initial fused matrix diagram to obtain a preset diagonal Markov matrix.

[0113] Iterative calculation subunit 2044 is used to replace the initial Mahalanobis identity matrix with a preset diagonal Mahalanobis matrix and trigger similarity calculation subunit 2041 until the iteration stop condition is reached, thus obtaining the fused similarity matrix graph.

[0114] Furthermore, the fusion analysis subunit 2042 is specifically used for:

[0115] Based on the similarity matrix, metric matrix view and preset penalty term, a fusion analysis of similarity is performed to obtain local structure fusion parameters;

[0116] Based on the local structure fusion parameters, a similar graph corresponding to the metric matrix view is constructed to obtain the initial fusion matrix graph.

[0117] This application also provides a consumer user classification device based on multi-view clustering, the device including a processor and memory;

[0118] The memory is used to store program code and transfer the program code to the processor;

[0119] The processor is used to execute the consumer user classification method based on multi-view clustering in the above method embodiments according to the instructions in the program code.

[0120] This application also provides a computer-readable storage medium for storing program code for executing the consumer user classification method based on multi-view clustering in the above method embodiments.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed apparatus 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 coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0122] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the various embodiments of this application through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0125] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for classifying consumer users based on multi-view clustering, characterized in that, The method comprises the following steps: feature extraction is performed on a multi-view consumption data set of consumers to obtain a multi-view data matrix; a K reverse neighbor algorithm is used to analyze reverse neighbor relationships between users based on the multi-view data matrix to obtain a K reverse neighbor data set; a Gaussian kernel similarity between any two user consumption operations in the K reverse neighbor data set is calculated to construct a metric matrix view; a preset diagonal Mahalanobis matrix is used to perform similarity fusion analysis on the metric matrix view to obtain a fused similarity matrix, and the specific process is as follows: a similarity Mahalanobis matrix between users in the multi-view data matrix is calculated based on an initial Mahalanobis unit matrix; a similarity fusion analysis is performed based on the similarity Mahalanobis matrix, the metric matrix view and a preset penalty term to obtain an initial fused matrix, and the specific process is as follows: a fusion analysis between similarities is performed based on the similarity Mahalanobis matrix, the metric matrix view and a preset penalty term to obtain a local structure fusion parameter; a similarity graph corresponding to the metric matrix view is constructed based on the local structure fusion parameter to obtain an initial fused matrix; a semi-positive definite Mahalanobis matrix update operation is performed based on the initial fused matrix to obtain a preset diagonal Mahalanobis matrix; the preset diagonal Mahalanobis matrix is replaced by the initial Mahalanobis unit matrix, and the step of calculating the similarity Mahalanobis matrix between users in the multi-view data matrix based on the initial Mahalanobis unit matrix is returned until an iteration stopping condition is reached to obtain a fused similarity matrix; a user clustering analysis is performed on the fused similarity matrix by Laplacian rank constraint to obtain a consumption user clustering result. 2.The multi-view clustering based consumer classification method of claim 1, wherein, The K reverse neighbor algorithm is used to analyze reverse neighbor relationships between users based on the multi-view data matrix to obtain a K reverse neighbor data set, and the method comprises the following steps: The K reverse neighbor algorithm is used to find, based on the multi-view data matrix, reverse neighbor samples of each user consumption operation sample in each consumption behavior to other user consumption operation samples of neighbor points to obtain a K reverse neighbor data set.

3. Apparatus for classifying consumer users based on multi-view clustering, characterized in that, The method comprises the following steps: a feature extraction unit is configured to perform feature extraction on a multi-view consumption data set of consumers to obtain a multi-view data matrix; a neighbor analysis unit is configured to use a K reverse neighbor algorithm to analyze reverse neighbor relationships between users based on the multi-view data matrix to obtain a K reverse neighbor data set; a similarity calculation unit is configured to calculate a Gaussian kernel similarity between any two user consumption operations in the K reverse neighbor data set to construct a metric matrix view; a similarity fusion unit is configured to use a preset diagonal Mahalanobis matrix to perform similarity fusion analysis on the metric matrix view to obtain a fused similarity matrix, and the similarity fusion unit comprises the following units: a similarity calculation subunit is configured to calculate a similarity Mahalanobis matrix between users in the multi-view data matrix based on an initial Mahalanobis unit matrix; a fusion analysis subunit is configured to perform similarity fusion analysis based on the similarity Mahalanobis matrix, the metric matrix view and a preset penalty term to obtain an initial fused matrix, and the fusion analysis subunit is specifically configured to: The fusion analysis between the similarities is performed according to the similar Mahalanobis matrix, the metric matrix view and a preset penalty term, to obtain a local structure fusion parameter; An initial fusion matrix view is obtained by constructing a corresponding similarity graph of the metric matrix view based on the local structure fusion parameter; A matrix updating subunit is configured to perform a semi-positive definite Mahalanobis matrix updating operation according to the initial fusion matrix view, to obtain a preset diagonal line Mahalanobis matrix; An iterative calculation subunit is configured to replace the initial Mahalanobis unit matrix with the preset diagonal line Mahalanobis matrix, and trigger the similarity calculation subunit until an iterative stop condition is reached, to obtain a fusion similarity matrix view; A clustering analysis unit is configured to perform user clustering analysis on the fusion similarity matrix view by Laplacian rank constraint, to obtain a consumption user clustering result.

4. The multi-view clustering based consumer classification apparatus of claim 3, wherein, The neighbor analysis unit is specifically configured to: find, according to the multi-view data matrix, a reverse neighbor sample of a neighbor point of each user consumption operation sample in each consumption behavior to other user consumption operation samples, to obtain a K reverse neighbor data set, based on a K reverse neighbor algorithm.

5. A device for classifying consumer users based on multi-view clustering, characterized by, The device comprises a processor and a memory; The memory is configured to store program code and transmit the program code to the processor; The processor is configured to execute the method according to any one of claims 1-2 based on instructions in the program code.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium is configured to store program code for executing the method according to any one of claims 1-2 based on multi-view clustering. The computer readable storage medium is configured to store program code for executing the method according to any one of claims 1-2 based on multi-view clustering.

Images