Consumer classification method and system based on multi-view clustering

Abstract

The invention discloses a consumption user classification method and system based on multi-view clustering, and is used for solving the technical problem of poor reliability of a classification result caused by an existing consumption user classification technology based on multi-view clustering. The method comprises the steps of obtaining multiple pieces of consumer behavior data, and performing similarity graph learning on the multiple pieces of consumer behavior data by adopting a preset similarity algorithm to generate multiple initial similarity graph matrixes; separating each initial similar graph matrix, and determining a consistency matrix and an inconsistency matrix corresponding to each initial similar graph matrix; according to the consistency matrix and the inconsistency matrix corresponding to each initial similar graph matrix, outputting an independence criterion and value and intra-view inter-view high-order information; respectively optimizing each initial similar graph matrix by adopting an independence criterion, a value and a consistency tensor, and determining a target similar graph matrix corresponding to each initial similar graph matrix; and generating a consumer classification result according to the plurality of target similar graph matrixes.

Description

Technical Field

[0001] The present invention relates to the technical field of cluster analysis, and in particular to a consumer user classification method and system based on multi-view clustering. Background Art

[0002] With the popularization of Internet technology, consumers have higher requirements for consumption experience, more diversified consumer goods, and higher consumption enthusiasm. In order to better meet the needs of consumers, consumer recommendation systems have become a key technology for major platforms to improve user experience and promote product sales. The recommendation system analyzes user historical behavior data, product attributes, social network data and other information, and uses cluster analysis to group target customers in order to recommend suitable products to customers and promote consumer consumption.

[0003] Multi-view clustering has long been widely used in many fields such as data mining and analysis, and is now attracting more and more attention. This is mainly due to the vigorous development of information technology, which has led to an explosive growth of multi-view data. When faced with massive data, clustering is one of the commonly used analysis techniques, and multi-view clustering is particularly important because a sample in multi-view data needs to be represented by features from multiple views. In real life, a sample often has different attributes and can be expressed in many ways. For example, objects on the Internet can be described in different ways such as photos, text, or video and voice; for consumers, it is necessary to analyze from five aspects: click to read, favorite, add to shopping cart, purchase, and after-sales service. After that, the feature consistency of multiple views is merged, and then consumers are grouped according to certain feature similarities.

[0004] Most of the existing consumer user classification technologies based on multi-view clustering use matrix-based methods or tensor-based methods for clustering, but these methods simply treat each consumer behavior as an independent feature and cannot accurately capture the full picture of user behavior patterns, resulting in poor reliability of classification results. Summary of the invention

[0005] The present invention provides a consumer user classification method and system based on multi-view clustering, which are used to solve the technical problem that the existing consumer user classification technology based on multi-view clustering leads to poor reliability of classification results.

[0006] A first aspect of the present invention provides a consumer user classification method based on multi-view clustering, comprising:

[0007] Acquire multiple consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the multiple consumer behavior data to generate multiple initial similarity graph matrices;

[0008] Separating each of the initial similarity graph matrices, and determining a consistency matrix and an inconsistency matrix corresponding to each of the initial similarity graph matrices;

[0009] Using a preset multi-view processing algorithm, outputting independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices;

[0010] The independence criteria and values ​​and the consistency tensor are used to optimize each of the initial similarity graph matrices, and a target similarity graph matrix corresponding to each of the initial similarity graph matrices is determined;

[0011] Generate a fused similarity graph matrix according to the plurality of target similarity graph matrices;

[0012] Perform spectral clustering on the fused similarity graph matrix to generate consumer user classification results.

[0013] Optionally, the preset multi-view processing algorithm includes an independence criterion and a double enhanced tensor rank minimization algorithm; the intra-view and inter-view high-order information includes inter-view high-order information and intra-view high-order information; the preset multi-view processing algorithm is used to output the independence criterion and value and the intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices, including:

[0014] Determine a plurality of independence criterion values ​​according to the inconsistency matrix corresponding to each of the initial similarity graph matrices using the independence criterion;

[0015] Summing the independence criterion values ​​to determine the independence criterion sum value;

[0016] Stacking the consistency matrices to generate a consistency stack tensor;

[0017] Rotating the consistent stacked tensor to determine a consistent rotated tensor;

[0018] A double-enhanced tensor rank minimization algorithm is used to determine the intra-view high-order information corresponding to the consistent stacked tensor and the inter-view high-order information corresponding to the consistent rotation tensor based on the consistent stacked tensor and the consistent rotation tensor.

[0019] Optionally, generating a fused similarity graph matrix according to the plurality of target similarity graph matrices comprises:

[0020] Separating each of the target similarity graph matrices, and determining a target consistency matrix corresponding to each of the target similarity graph matrices;

[0021] A plurality of the target consistency matrices are fused to generate a fused similarity graph matrix.

[0022] Optionally, the construction process of the initial similarity graph matrix is ​​specifically as follows:

[0023] ;

[0024] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization.

[0025] Optionally, the separation process of the initial similarity graph matrix is ​​specifically as follows:

[0026] ;

[0027] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; is a trade-off parameter.

[0028] Optionally, the optimization process of the initial similarity graph matrix is ​​specifically as follows:

[0029] ;

[0030] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; For trade-off parameters; is an adjustable parameter; is the independence criterion and value; For the principle of independence; It is the high-level information in the view; It is the high-level information between views; For stacking; is a rotation; A is a consistent stacking tensor; Rotate the tensor for consistency.

[0031] A second aspect of the present invention provides a consumer user classification system based on multi-view clustering, comprising:

[0032] An acquisition module, used to acquire a plurality of consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the plurality of consumer behavior data to generate a plurality of initial similarity graph matrices;

[0033] A separation module, used to separate each of the initial similarity graph matrices, and determine a consistency matrix and an inconsistency matrix corresponding to each of the initial similarity graph matrices;

[0034] An output module, configured to output inconsistency independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices by using a preset multi-view processing algorithm;

[0035] An optimization module, used to optimize each of the initial similarity graph matrices respectively by using the inconsistency independence criterion and value and the consistency tensor, and determine a target similarity graph matrix corresponding to each of the initial similarity graph matrices;

[0036] A generating module, used for generating a fused similarity graph matrix according to a plurality of target similarity graph matrices;

[0037] The spectral clustering module is used to perform spectral clustering on the fused similarity graph matrix to generate a consumer user classification result.

[0038] A third aspect of the present invention provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the consumer user classification method based on multi-view clustering as described in any one of the above items.

[0039] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the steps of the consumer user classification method based on multi-view clustering as described in any one of the above are implemented.

[0040] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the consumer user classification method based on multi-view clustering as described in any one of the above items.

[0041] It can be seen from the above technical solutions that the present invention has the following advantages:

[0042] The above technical scheme of the present invention provides a consumer user classification method based on multi-view clustering, which first obtains multiple consumer behavior data, and uses a preset similarity algorithm to perform similarity graph learning on the multiple consumer behavior data to generate multiple initial similarity graph matrices; then, separates each initial similarity graph matrix to determine the consistency matrix and inconsistency matrix corresponding to each initial similarity graph matrix; uses a preset multi-view processing algorithm to output independence criteria and values ​​and high-order information between views within a view according to the consistency matrix and inconsistency matrix corresponding to each initial similarity graph matrix; uses the independence criteria and values ​​and consistency tensor to optimize each initial similarity graph matrix respectively, and determines the target similarity graph matrix corresponding to each initial similarity graph matrix; based on multiple target similarity graph matrices, Generate a fused similarity graph matrix; finally, perform spectral clustering on the fused similarity graph matrix to generate consumer user classification results; based on the above scheme, combined with a preset multi-view processing algorithm and a preset similarity algorithm, multiple consumer behavior data are processed, and independence criteria and values ​​and high-order information between views within a view are output, and each initial similarity graph matrix is ​​optimized using the independence criteria and values ​​and high-order information between views within a view, and the target similarity graph matrix corresponding to each initial similarity graph matrix is ​​determined, and then spectral clustering is performed on the fused similarity graph matrix generated according to multiple target similarity graph matrices to generate the consumer user classification results. The process can better consider higher-order correlations between and within behaviors, thereby better grouping consumers, thereby improving the reliability of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A flowchart of a method for classifying consumer users based on multi-view clustering provided in the first embodiment of the present invention;

[0045] Figure 2 This is an overall framework diagram of the consumer user classification method based on multi-view clustering provided in the first embodiment of the present invention;

[0046] Figure 3 A schematic diagram of a process of a consumer user classification method based on multi-view clustering provided in the second embodiment of the present invention;

[0047] Figure 4 This is a structural block diagram of a consumer user classification system based on multi-view clustering provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0048] The embodiments of the present invention provide a method and system for classifying consumer users based on multi-view clustering, which are used to solve the technical problem that the existing consumer user classification technology based on multi-view clustering leads to poor reliability of classification results.

[0049] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 , Figure 1 A flowchart of the steps of a consumer user classification method based on multi-view clustering provided in the first embodiment of the present invention.

[0051] The present invention provides a consumer user classification method based on multi-view clustering, comprising:

[0052] Step 101: Acquire multiple consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the multiple consumer behavior data to generate multiple initial similarity graph matrices.

[0053] The multiple consumer behavior data include multiple consumers' reading, collection, adding to shopping cart, purchase, after-sales and other behavior data.

[0054] It should be noted that the preset similarity algorithm uses Euclidean distance learning similarity to learn the similarity matrix, thereby obtaining the initial similarity graph matrix corresponding to each initial similarity graph. This is also a commonly used similarity calculation method in multi-view clustering. The construction process of the initial similarity graph matrix is ​​specifically as follows:

[0055] ;

[0056] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization.

[0057] Step 102: Separate each initial similarity graph matrix and determine a consistency matrix and an inconsistency matrix corresponding to each initial similarity graph matrix.

[0058] It should be noted that the same consumer may have some unique records when performing different behaviors, which will cause the similarity matrix S to contain incorrect sample similarities. Therefore, for the accuracy of the results, the present invention will separate these special records for processing. Therefore, the present invention separates the initial similarity graph matrix S into a consistent part A (consistency matrix) and an inconsistent part E (inconsistency matrix). The inconsistency here is a broader definition than noise; the separation process of the initial similarity graph matrix is ​​specifically as follows:

[0059] ;

[0060] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; is a trade-off parameter.

[0061] Step 103: Use a preset multi-view processing algorithm to output independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each initial similarity graph matrix.

[0062] The preset multi-view processing algorithms include the independence criterion and the double enhanced tensor rank minimization algorithm. The independence criterion is the Hilbert-Schmidt Independence Criterion (HSIC).

[0063] The intra-view and inter-view high-order information includes inter-view high-order information and intra-view high-order information.

[0064] Specifically, step 103 may include the following sub-steps S31-S35:

[0065] Step S31, using the independence criterion to determine a plurality of independence criterion values ​​according to the inconsistency matrix corresponding to each initial similarity graph matrix;

[0066] It should be noted that in order to mine the complementary information between behaviors, the Hilbert-Schmidt independence criterion is used to enhance the diversity of the inconsistent parts. The Hilbert-Schmidt independence criterion (HSIC) is a method for detecting the independence between random variables. It is based on the theory of kernel methods and determines whether they are independent by calculating the degree of dependence between random variables. If two variables are independent, then their covariance should be zero.

[0067] Furthermore, the processing principle of the Hilbert-Schmidt independence criterion is as follows: given two random variables X and Y, choose an appropriate sum function k X and k Y , in order to map X and Y into a high-dimensional feature space and calculate the kernel matrix K of X and Y X and K Y, and then each kernel matrix is ​​centered to eliminate the influence of the mean, and finally the HSIC value is calculated. The closer the HSIC value is to zero, the more independent X and Y are; the larger the HSIC value is, the stronger the dependence between them is.

[0068] Step S32, summing up the independence criterion values ​​to determine the independence criterion sum value;

[0069] Step S33, stacking each consistency matrix to generate a consistency stack tensor;

[0070] Step S34, rotating the consistent stacked tensor to determine a consistent rotated tensor;

[0071] Step S35: Using a double-enhanced tensor rank minimization algorithm, based on the consistent stacking tensor and the consistent rotation tensor, determine the intra-view high-order information corresponding to the consistent stacking tensor and the inter-view high-order information corresponding to the consistent rotation tensor.

[0072] It should be noted that, based on the diversified separation of inconsistent information, the present invention utilizes Figure 1 The high-order information of the consistent part is used to improve the performance of the model. However, the tensor nuclear norm, which is usually used to explore the high-order information of the view, has the same penalty for different singular values ​​and cannot approximate the true rank of the tensor well. Therefore, the present invention introduces a double enhanced tensor rank minimization algorithm, and the definition of the enhanced tensor rank is as follows:

[0073] ;

[0074] in, To enhance the tensor rank, it represents high-order information; is the dimension of the third dimension of the tensor; To represent the tensor P f The b-th frontal slice of ; is an adjustable parameter, ; For tensors Singular values ​​obtained after tensor singular value decomposition in the Fourier domain; is an adjustable parameter; t is a square matrix The dimension of , for The dimension of the first dimension of for The dimensionality of the second dimension.

[0075] Furthermore, in order to simultaneously obtain high-level information between views and within views, the present invention stacks all consistency matrices to obtain a tensor A, and rotates the tensor A to obtain , tensor rank minimization (double enhanced tensor rank minimization algorithm) is used to process A and , then A can be used to mine high-level information within the attempt. Used to mine high-level information between views.

[0076] Step 104: Use the independence criterion and value and the consistency tensor to optimize each initial similarity graph matrix respectively, and determine the target similarity graph matrix corresponding to each initial similarity graph matrix.

[0077] It should be noted that the optimization process of the initial similarity graph matrix is ​​as follows:

[0078] ;

[0079] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; For trade-off parameters; is an adjustable parameter; is the independence criterion and value; For the principle of independence; It is the high-level information in the view; It is the high-level information between views; For stacking; is a rotation; A is a consistent stacking tensor; Rotate the tensor for consistency.

[0080] Step 105: Generate a fused similarity graph matrix based on multiple target similarity graph matrices.

[0081] It should be noted that the consistent parts of all target similarity graph matrices are fused to obtain a unified fused similarity graph matrix Z.

[0082] Step 106: Perform spectral clustering on the fused similarity graph matrix to generate consumer user classification results.

[0083] It should be noted that spectral clustering is performed on Z to obtain clustering results (consumer user classification results).

[0084] For example, see Figure 2 First, a preset similarity algorithm is used to process multiple input consumer behavior data to obtain multiple initial similarity graph matrices, and each initial similarity graph matrix is ​​separated to determine the consistency matrix and inconsistency matrix corresponding to each initial similarity graph matrix. Then, combined with the preset multi-view processing algorithm, all consistency matrices and inconsistency matrices are processed to output the independence criterion and value and the high-order information between views within the view. The independence criterion and value and the high-order information between views within the view are used to optimize each initial similarity graph matrix to determine the target similarity graph matrix corresponding to each initial similarity graph matrix. Finally, the fused similarity graph matrix generated according to multiple target similarity graph matrices is spectrally clustered to generate the consumer user classification result.

[0085] As a comparison of technical effects, it can be combined with existing technologies for reference. For multi-view data, many methods have been proposed to improve the performance of multi-view clustering. Existing multi-view clustering methods can be roughly divided into matrix-based methods and tensor-based methods. Matrix-based methods can also be divided into subspace learning, matrix decomposition, graph learning, etc. Among these methods, graph-based clustering methods can well learn the similarity relationship between samples and achieve good clustering effects. At the same time, tensor-based clustering methods have become a hot direction for multi-view clustering due to their excellent performance in capturing high-order information of views. However, the above methods have the following problems: 1) Failure to pay attention to the diversity of inconsistent information between consumer behaviors, resulting in poor performance; 2) Failure to consider higher-order correlations between and within behaviors; 3) The same penalty is taken for different singular values ​​of the tensor.

[0086] In view of the above problems, the present invention proposes a consumer user classification method based on multi-view clustering. By analyzing various behaviors of consumers and using a multi-view clustering method based on tensor rank minimization and consideration of view inconsistency, consumers can be better grouped and suitable products can be recommended to different consumer groups. Specifically, consumers' reading, collecting, adding to shopping carts, purchasing, after-sales and other behaviors are sampled, and the initial similarity graph of each behavior (i.e., view) is extracted, and the inconsistency factor between views is considered. The inconsistency may be noise caused by consumer erroneous operation, or it may be a view-specific feature. Enhanced tensor rank minimization is used to enhance the high-order correlation of consistency information between and within views, the consistency tensors of each view are fused into a graph, and the fused graph is spectrally clustered to achieve consumer grouping, and suitable products are recommended to consumers based on related searches.

[0087] In an embodiment of the present invention, the present invention provides a consumer user classification method based on multi-view clustering, firstly, a plurality of consumer behavior data are obtained, and a preset similarity algorithm is used to perform similarity graph learning on the plurality of consumer behavior data to generate a plurality of initial similarity graph matrices; then, each initial similarity graph matrix is ​​separated to determine the consistency matrix and the inconsistency matrix corresponding to each initial similarity graph matrix; a preset multi-view processing algorithm is used to output the independence criterion and value and the high-order information between views within a view according to the consistency matrix and the inconsistency matrix corresponding to each initial similarity graph matrix; the independence criterion and value and the consistency tensor are used to optimize each initial similarity graph matrix respectively to determine the target similarity graph matrix corresponding to each initial similarity graph matrix; according to the plurality of target similarity graph matrices , generate a fused similarity graph matrix; finally, perform spectral clustering on the fused similarity graph matrix to generate consumer user classification results; based on the above scheme, combined with the preset multi-view processing algorithm and the preset similarity algorithm, multiple consumer behavior data are processed, and the independence criteria and values ​​and high-order information between views within a view are output, and the independence criteria and values ​​and high-order information between views within a view are used to optimize each initial similarity graph matrix, determine the target similarity graph matrix corresponding to each initial similarity graph matrix, and then perform spectral clustering on the fused similarity graph matrix generated according to multiple target similarity graph matrices to generate the consumer user classification result. The process can better consider the higher-order correlations between and within behaviors, thereby better grouping consumers, thereby improving the reliability of the classification results.

[0088] For better explanation, refer to Figure 3 , shows a flow chart of a consumer user classification method based on multi-view clustering provided in Embodiment 2 of the present invention, including:

[0089] First, based on the input consumer behavior data, each initial similarity graph matrix is ​​learned. Then, the initial similarity graph matrix S is separated into a consistent part and an inconsistent part, and then all the consistent parts are superimposed into a three-dimensional tensor A. It can be obtained by rotating the tensor A. Then the tensor is subjected to enhanced tensor rank minimization processing and fed back to the initial similarity graph matrix for optimization. Finally, the consistent parts of all optimized similarity matrices (target similarity graph matrices) are fused to obtain the fused similarity graph matrix Z, and spectral clustering is performed on Z to obtain the clustering result.

[0090] See also Figure 4 , Figure 4 This is a structural block diagram of a consumer user classification system based on multi-view clustering provided in Example 3 of the present invention.

[0091] The present invention provides a consumer user classification system based on multi-view clustering, comprising:

[0092] The acquisition module 401 is used to acquire multiple consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the multiple consumer behavior data to generate multiple initial similarity graph matrices;

[0093] A separation module 402 is used to separate each initial similarity graph matrix and determine a consistency matrix and an inconsistency matrix corresponding to each initial similarity graph matrix;

[0094] An output module 403 is used to output inconsistency independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each initial similarity graph matrix by using a preset multi-view processing algorithm;

[0095] An optimization module 404 is used to optimize each initial similarity graph matrix using an inconsistency independence criterion and value and a consistency tensor, and determine a target similarity graph matrix corresponding to each initial similarity graph matrix;

[0096] A generating module 405 is used to generate a fused similarity graph matrix according to multiple target similarity graph matrices;

[0097] The spectral clustering module 406 is used to perform spectral clustering on the fused similarity graph matrix to generate a consumer user classification result.

[0098] Further, the preset multi-view processing algorithm includes an independence criterion and a double enhanced tensor rank minimization algorithm; the intra-view and inter-view high-order information includes inter-view high-order information and intra-view high-order information; the output module 403 is specifically used to:

[0099] Using the independence criterion, a plurality of independence criterion values ​​are determined according to the inconsistency matrix corresponding to each initial similarity graph matrix;

[0100] Sum up the values ​​of each independence criterion to determine the independence criterion and value;

[0101] Stack each consistency matrix to generate a consistency stack tensor;

[0102] Rotate the consistent stacked tensor to determine a consistent rotated tensor;

[0103] A double-augmented tensor rank minimization algorithm is used to determine the intra-view high-order information corresponding to the consistent stacking tensor and the inter-view high-order information corresponding to the consistent rotation tensor based on the consistent stacking tensor and the consistent rotation tensor.

[0104] Furthermore, the generating module 405 is specifically used for:

[0105] Separate each target similarity graph matrix and determine the target consistency matrix corresponding to each target similarity graph matrix;

[0106] Multiple target consistency matrices are fused to generate a fused similarity graph matrix.

[0107] Furthermore, the construction process of the initial similarity graph matrix is ​​as follows:

[0108] ;

[0109] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization.

[0110] Furthermore, the separation process of the initial similarity graph matrix is ​​as follows:

[0111] ;

[0112] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; is a trade-off parameter.

[0113] Furthermore, the optimization process of the initial similarity graph matrix is ​​as follows:

[0114] ;

[0115] in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the j-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; For trade-off parameters; is an adjustable parameter; is the independence criterion and value; For the principle of independence; It is the high-level information in the view; It is the high-level information between views; For stacking; is a rotation; A is a consistent stacking tensor; Rotate the tensor for consistency.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system and module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0117] An embodiment of the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the consumer user classification method based on multi-view clustering as in any of the above embodiments.

[0118] An embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the consumer user classification method based on multi-view clustering as in any of the above embodiments are implemented.

[0119] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the consumer user classification method based on multi-view clustering as in any of the above embodiments.

[0120] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A consumer user classification method based on multi-view clustering, characterized in that: include: Acquire multiple consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the multiple consumer behavior data to generate multiple initial similarity graph matrices; Separating each of the initial similarity graph matrices, and determining a consistency matrix and an inconsistency matrix corresponding to each of the initial similarity graph matrices; Using a preset multi-view processing algorithm, outputting independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices; The independence criteria and values ​​and the consistency tensor are used to optimize each of the initial similarity graph matrices, and a target similarity graph matrix corresponding to each of the initial similarity graph matrices is determined; Generate a fused similarity graph matrix according to the plurality of target similarity graph matrices; Perform spectral clustering on the fused similarity graph matrix to generate consumer user classification results.

2. The method for classifying consumer users based on multi-view clustering according to claim 1, characterized in that: The preset multi-view processing algorithm includes an independence criterion and a double enhanced tensor rank minimization algorithm; the intra-view and inter-view high-order information includes inter-view high-order information and intra-view high-order information; the preset multi-view processing algorithm is used to output the independence criterion and value and the intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices, including: Determine a plurality of independence criterion values ​​according to the inconsistency matrix corresponding to each of the initial similarity graph matrices using the independence criterion; Summing the independence criterion values ​​to determine the independence criterion sum value; Stacking the consistency matrices to generate a consistency stack tensor; Rotating the consistent stacked tensor to determine a consistent rotated tensor; A double-enhanced tensor rank minimization algorithm is used to determine the intra-view high-order information corresponding to the consistent stacked tensor and the inter-view high-order information corresponding to the consistent rotation tensor based on the consistent stacked tensor and the consistent rotation tensor.

3. The method for classifying consumer users based on multi-view clustering according to claim 1, characterized in that: The step of generating a fused similarity graph matrix according to the plurality of target similarity graph matrices comprises: Separating each of the target similarity graph matrices, and determining a target consistency matrix corresponding to each of the target similarity graph matrices; A plurality of the target consistency matrices are fused to generate a fused similarity graph matrix.

4. The method for classifying consumer users based on multi-view clustering according to claim 1, characterized in that: The construction process of the initial similarity graph matrix is ​​specifically as follows: ； in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the jth consumer behavior data corresponding to the vth initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization.

5. The method for classifying consumer users based on multi-view clustering according to claim 1, characterized in that: The separation process of the initial similarity graph matrix is ​​specifically as follows: ； in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the jth consumer behavior data corresponding to the vth initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; is a trade-off parameter.

6. The method for classifying consumer users based on multi-view clustering according to claim 1, characterized in that: The optimization process of the initial similarity graph matrix is ​​specifically as follows: ； in, is the vth initial similarity graph matrix; m is the total number of initial similarity graph matrices; n is the total number of samples of consumer behavior data; is the initial similarity value corresponding to the vth initial similarity graph matrix composed of the i-th consumer behavior data and the j-th consumer behavior data; is the i-th consumer behavior data corresponding to the v-th initial similarity graph matrix; is the jth consumer behavior data corresponding to the vth initial similarity graph matrix; is the square of the Euclidean norm; T is the transpose; For Perform regularization; is the consistency matrix corresponding to the vth initial similarity graph matrix; is the inconsistency matrix corresponding to the vth initial similarity graph matrix; For trade-off parameters; is an adjustable parameter; is the independence criterion and value; For the principle of independence; It is the high-level information in the view; It is the high-level information between views; For stacking; is a rotation; A is a consistent stacking tensor; Rotate the tensor for consistency.

7. A consumer user classification system based on multi-view clustering, characterized in that: include: An acquisition module, used to acquire a plurality of consumer behavior data, and use a preset similarity algorithm to perform similarity graph learning on the plurality of consumer behavior data to generate a plurality of initial similarity graph matrices; A separation module, used to separate each of the initial similarity graph matrices, and determine a consistency matrix and an inconsistency matrix corresponding to each of the initial similarity graph matrices; An output module, configured to output inconsistency independence criteria and values ​​and intra-view and inter-view high-order information according to the consistency matrix and the inconsistency matrix corresponding to each of the initial similarity graph matrices by using a preset multi-view processing algorithm; An optimization module, used to optimize each of the initial similarity graph matrices respectively by using the inconsistency independence criterion and value and the consistency tensor, and determine a target similarity graph matrix corresponding to each of the initial similarity graph matrices; A generating module, used for generating a fused similarity graph matrix according to a plurality of target similarity graph matrices; The spectral clustering module is used to perform spectral clustering on the fused similarity graph matrix to generate a consumer user classification result.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the consumer user classification method based on multi-view clustering as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the consumer user classification method based on multi-view clustering as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the consumer user classification method based on multi-view clustering as described in any one of claims 1-6.

Images