User tag prediction method and device

Abstract

The invention provides a user tag prediction method and device, relates to the field of artificial intelligence, and can be used in the financial field or other fields. The method comprises the steps of obtaining a behavior transaction information table authorized by a user, performing key information extraction on the behavior transaction information table to obtain multiple pieces of user feature information, and determining a feature missing rate corresponding to each piece of user feature information; comparing a preset missing rate threshold with the feature missing rate, performing data cleaning on the user feature information, and performing data recovery on the user feature information after data cleaning by using a sparse subspace algorithm; and processing the user feature information after data recovery by using a pre-established prediction model to obtain a user tag prediction result. According to the method, the characteristics and behaviors of the user can be described more accurately, the personalized requirements of clients are better met, the data accuracy is improved, the label division efficiency is improved, accurate division of user labels is realized, the user experience is improved, the resource waste is reduced, and the popularization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a user tag prediction method and device. Background Art

[0002] At present, when banks acquire new customers, they can only label customers based on the simple information provided by the customers. In this case, they can only roughly label customers based on the current transactions made by the customers and the simple record of customer portraits. Obviously, there are problems such as low work efficiency, inaccurate user labeling, and waste of human and material resources. Therefore, there is no mature solution for how to establish accurate customer labels for new customers. How to improve the accuracy of user labeling, improve the efficiency of labeling, and save human and material resources has become an urgent problem to be solved. Summary of the invention

[0003] In view of the problems existing in the prior art, the main purpose of the embodiments of the present invention is to provide a user tag prediction method and device to achieve accurate prediction of user tags.

[0004] In order to achieve the above object, an embodiment of the present invention provides a user tag prediction method, the method comprising:

[0005] Obtaining a behavior transaction information table authorized by the user, extracting key information from the behavior transaction information table, obtaining multiple user feature information, and determining a feature missing rate corresponding to each user feature information;

[0006] Comparing a preset missing rate threshold with the feature missing rate, performing data cleaning on the user feature information, and using a sparse subspace algorithm to perform data recovery on the user feature information after data cleaning;

[0007] The pre-established prediction model is used to process the user feature information after data recovery to obtain the user label prediction result.

[0008] Optionally, in an embodiment of the present invention, determining the feature missing rate corresponding to each user feature information includes:

[0009] Count the number of missing feature categories of user feature information of a single user, and the total number of feature categories of user feature information of all users;

[0010] The feature missing rate corresponding to each user feature information is determined according to the number of missing feature categories and the total number of feature categories.

[0011] Optionally, in an embodiment of the present invention, comparing a preset missing rate threshold with the feature missing rate, and performing data cleaning on the user feature information includes:

[0012] The preset missing rate threshold is compared with the feature missing rate, and the user feature information whose feature missing rate is greater than the preset missing rate threshold is eliminated to complete the data cleaning of the user feature information.

[0013] Optionally, in an embodiment of the present invention, using a sparse subspace algorithm to recover the user characteristic information after data cleaning includes:

[0014] According to the user characteristic information after data cleaning, determine the behavior characteristic data matrix and the self-representation matrix;

[0015] According to the behavior feature data matrix and the self-representation matrix, a sparse subspace algorithm is used to recover the user feature information after data cleaning to obtain complete user feature data.

[0016] Optionally, in an embodiment of the present invention, processing the user characteristic information after data recovery by using a pre-established prediction model includes:

[0017] Determine a feature extraction matrix corresponding to the complete feature data of the user according to the complete feature data of the user;

[0018] The feature extraction matrix is ​​processed using the preset independence indicator calculation model to obtain the user behavior transaction matrix.

[0019] Optionally, in an embodiment of the present invention, processing the user characteristic information after data recovery by using a pre-established prediction model further includes:

[0020] The user behavior transaction matrix is ​​predicted using a pre-established multi-label classification model, wherein the multi-label classification model includes a label co-occurrence manifold regularization term.

[0021] Optionally, in one embodiment of the present invention, the method further includes:

[0022] Obtain a sample data set with some missing labels and its corresponding label matrix, and use the sample data set and its corresponding label matrix to determine the label co-occurrence similarity;

[0023] According to the Laplace manifold learning algorithm, the label co-occurrence similarity is used to construct the manifold regularization term, and the label co-occurrence manifold regularization term is obtained.

[0024] The embodiment of the present invention further provides a user tag prediction device, the device comprising:

[0025] A feature missing rate module is used to obtain a behavior transaction information table authorized by a user, extract key information from the behavior transaction information table, obtain multiple user feature information, and determine a feature missing rate corresponding to each user feature information;

[0026] A data recovery module, used to compare a preset missing rate threshold with the feature missing rate, perform data cleaning on the user feature information, and use a sparse subspace algorithm to perform data recovery on the user feature information after data cleaning;

[0027] The label prediction module is used to process the user feature information after data recovery using a pre-established prediction model to obtain a user label prediction result.

[0028] Optionally, in one embodiment of the present invention, the feature missing rate module includes:

[0029] A feature category number unit is used to count the number of missing feature categories of user feature information of a single user, and the total number of feature categories of user feature information of all users;

[0030] The feature missing rate unit is used to determine the feature missing rate corresponding to each user feature information according to the number of missing feature categories and the total number of feature categories.

[0031] Optionally, in one embodiment of the present invention, the data recovery module is also used to compare a preset missing rate threshold with the feature missing rate, and eliminate user feature information whose feature missing rate is greater than the preset missing rate threshold to complete data cleaning of the user feature information.

[0032] Optionally, in one embodiment of the present invention, the data recovery module includes:

[0033] A self-representation matrix unit, used to determine a behavior feature data matrix and a self-representation matrix according to user feature information after data cleaning;

[0034] The data recovery unit is used to recover the user characteristic information after data cleaning based on the behavior characteristic data matrix and the self-representation matrix by using the sparse subspace algorithm to obtain the user's complete characteristic data.

[0035] Optionally, in one embodiment of the present invention, the label prediction module includes:

[0036] A feature extraction matrix unit, used to determine a feature extraction matrix corresponding to the complete feature data of the user according to the complete feature data of the user;

[0037] The behavior transaction matrix unit is used to process the feature extraction matrix using a preset independence indicator calculation model to obtain a user behavior transaction matrix.

[0038] Optionally, in one embodiment of the present invention, the label prediction module is also used to use a pre-established multi-label classification model to perform prediction processing on the user behavior transaction matrix; wherein the multi-label classification model includes a label co-occurrence manifold regularization term.

[0039] Optionally, in one embodiment of the present invention, the device further includes:

[0040] The co-occurrence similarity module is used to obtain the sample data set with some missing labels and its corresponding label matrix, and determine the label co-occurrence similarity by using the sample data set and its corresponding label matrix;

[0041] The manifold regularization term module is used to construct a manifold regularization term using label co-occurrence similarity according to a Laplace manifold learning algorithm to obtain the label co-occurrence manifold regularization term.

[0042] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above method is implemented when the processor executes the program.

[0043] The present invention also provides a computer-readable storage medium, which stores a computer program for executing the above method by a computer.

[0044] The present invention also provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when the computer program / instruction is executed by a processor.

[0045] The present invention calculates the feature missing rate of user feature information, cleans and restores user feature information, effectively completes missing information, improves data accuracy, improves the efficiency of label division, saves manpower and material resources, and combines the method of predicting user labels with a pre-established prediction model. It can more accurately describe user characteristics and behaviors, better meet customer personalized needs, achieve accurate division of user labels, improve user experience, reduce resource waste, and improve promotion efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] 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 description of the embodiments 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 work.

[0047] Figure 1 A flowchart of a user tag prediction method according to an embodiment of the present invention;

[0048] Figure 2 This is a flow chart of determining a feature missing rate in an embodiment of the present invention;

[0049] Figure 3 A flowchart of data recovery in an embodiment of the present invention;

[0050] Figure 4This is a flowchart of the prediction model data processing in an embodiment of the present invention;

[0051] Figure 5 A flowchart of obtaining a label co-occurrence manifold regularization term in an embodiment of the present invention;

[0052] Figure 6 A flowchart of a user tag prediction method in a specific embodiment of the present invention;

[0053] Figure 7 A schematic diagram of the structure of a user tag prediction device according to an embodiment of the present invention;

[0054] Figure 8 Schematic diagram of the structure of the feature missing rate module in an embodiment of the present invention;

[0055] Fig. 9 This is a schematic diagram of the structure of a data recovery module in an embodiment of the present invention;

[0056] Fig.10 This is a schematic diagram of the structure of a label prediction module in an embodiment of the present invention;

[0057] Fig.11 It is a structural schematic diagram of a user tag prediction device in another embodiment of the present invention;

[0058] Fig.12 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The embodiments of the present invention provide a user label prediction method and device, which can be used in the financial field and other fields. It should be noted that the user label prediction method and device of the present invention can be used in the financial field, and can also be used in any field other than the financial field. The application field of the user label prediction method and device of the present invention is not limited.

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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.

[0061] like Figure 1The flowchart of a user tag prediction method according to an embodiment of the present invention is shown. The execution subject of the user tag prediction method provided by the embodiment of the present invention includes but is not limited to a computer. The present invention calculates the feature missing rate of user feature information, cleans and restores user feature information, effectively completes missing information, improves data accuracy, improves the efficiency of tag division, saves human and material resources, and combines the method of predicting user tags with a pre-established prediction model. It can more accurately describe user characteristics and behaviors, better meet customer personalized needs, improve the precise division of user tags, improve user experience, reduce resource waste, and improve promotion efficiency. The method shown in the figure includes:

[0062] Step S1, obtaining a behavior transaction information table authorized by the user, extracting key information from the behavior transaction information table, obtaining multiple user feature information, and determining a feature missing rate corresponding to each user feature information.

[0063] Among them, after the user's authorization, the behavior transaction information table is obtained, and specifically, the behavior transaction information table includes the behavior transaction information of multiple users. In addition, the behavior transaction information table can be obtained in a conventional manner, which will not be repeated here.

[0064] Furthermore, key information is extracted from the behavior transaction information table to obtain multiple user feature information. Specifically, the user feature information may include user transaction information, such as transaction type, transaction frequency, etc. The corresponding feature missing rate can be determined through the user feature information. Specifically, by presetting the number of features that each user feature information should contain, it is possible to quickly determine whether the user feature information is missing a feature and its corresponding feature missing rate.

[0065] Step S2, comparing the preset missing rate threshold with the feature missing rate, performing data cleaning on the user feature information, and using a sparse subspace algorithm to restore the user feature information after data cleaning.

[0066] The preset missing rate value may be set to 0.5. When the feature missing rate of a certain user feature information is greater than 0.5, the user feature information is removed to complete data cleaning.

[0067] Furthermore, when the feature missing rate is less than the preset missing rate threshold, the sparse subspace algorithm is used, that is, the sample data sparse subspace self-representation model (SSC) is used to restore the missing features. The model is as follows:

[0068]

[0069] Among them, X represents the user's behavior feature data matrix, and C represents the self-representation matrix of the sample data, which is used to restore the sample features. Considering the error of user data, the noise matrix E is introduced and the l1 norm is used to constrain it. and o are model hyperparameters to prevent model overfitting, and finally obtain the complete feature data XC of the user. Thus, the data recovery process of the user feature information after data cleaning is completed.

[0070] Step S3, using the pre-established prediction model to process the user feature information after data recovery to obtain a user tag prediction result.

[0071] Among them, the present invention takes the classic multi-label classification model (LSML) as the basic model, and uses the label co-occurrence similarity measurement method to reconstruct the Laplace manifold regularization term, and uses the obtained label co-occurrence manifold regularization term as the last term in the LSML model, thereby obtaining a prediction model, which improves the prediction accuracy of the LSML model.

[0072] Furthermore, the user feature information after data recovery is input into the prediction model for processing to obtain the user tag prediction result. In addition, the user tag prediction result is used to construct a user portrait, thereby obtaining a richer user portrait and better understanding user needs and behaviors.

[0073] As an embodiment of the present invention, Figure 2 As shown, determining the feature missing rate corresponding to each user feature information includes:

[0074] Step S11, counting the number of missing feature categories of user feature information of a single user and the total number of feature categories of user feature information of all users;

[0075] Step S12: determining the feature missing rate corresponding to each user feature information according to the number of missing feature categories and the total number of feature categories.

[0076] The number of missing feature categories in each user feature information is determined by statistics, and the total number of feature categories of the user feature information of all users is counted. The ratio of the number of missing feature categories in a certain user feature information to the total number of feature categories is taken as the feature missing rate corresponding to the user feature information.

[0077] As an embodiment of the present invention, a preset missing rate threshold is compared with a feature missing rate, and data cleaning of user feature information includes: comparing the preset missing rate threshold with the feature missing rate, and eliminating user feature information whose feature missing rate is greater than the preset missing rate threshold, so as to complete data cleaning of user feature information.

[0078] The preset missing rate value may be set to 0.5. When the feature missing rate of a certain user feature information is greater than 0.5, the user feature information is removed to complete data cleaning.

[0079] As an embodiment of the present invention, Figure 3 As shown, using the sparse subspace algorithm, data recovery of user feature information after data cleaning includes:

[0080] Step S21, determining a behavior feature data matrix and a self-representation matrix according to the user feature information after data cleaning;

[0081] Step S22, based on the behavior feature data matrix and the self-representation matrix, using a sparse subspace algorithm, recover the user feature information after data cleaning to obtain complete user feature data.

[0082] Among them, the user characteristic information is used to construct the behavior characteristic data matrix and the self-representation matrix. The matrix construction process can be carried out in a conventional manner and will not be described in detail here.

[0083] Furthermore, the obtained behavior feature data matrix and the self-representation matrix are substituted into formula (1), and the sparse subspace algorithm is used to recover the user feature information after data cleaning to obtain the user's complete feature data.

[0084] In this embodiment, if Figure 4 As shown, using the pre-established prediction model to process the user characteristic information after data recovery includes:

[0085] Step S31, determining a feature extraction matrix corresponding to the user's complete feature data according to the user's complete feature data;

[0086] Step S32, using a preset independence index calculation model to process the feature extraction matrix to obtain a user behavior transaction matrix.

[0087] Among them, the present invention uses the Hilbert-Schmidt independence index (HSIC) to maximize the dependency of user tags and features to achieve feature extraction of user complete feature data. The specific model algorithm is as follows:

[0088]

[0089] Among them, P is a feature extraction matrix, i.e., a feature selection matrix, which is used to mine important features related to label information in user transaction behaviors. At the same time, the data dimension of the user behavior transaction matrix PX after feature screening is lower than the dimension of the original user transaction feature data X, and the features under PX have the greatest correlation with the user label information.

[0090] In this embodiment, using a pre-established prediction model to process user feature information after data recovery also includes: using a pre-established multi-label classification model to predict the user behavior transaction matrix; wherein the multi-label classification model includes a label co-occurrence manifold regularization term.

[0091] Among them, the label co-occurrence similarity measurement method is used to reconstruct the Laplace manifold regularization term, and the obtained label co-occurrence manifold regularization term is used as the last term in the multi-label classification model (LSML) model, thereby obtaining a prediction model, which improves the prediction accuracy of the multi-label classification model.

[0092] In this embodiment, if Figure 5 As shown, the method also includes:

[0093] Step S4, obtaining a sample data set with some missing labels and its corresponding label matrix, and using the sample data set and its corresponding label matrix to determine the label co-occurrence similarity;

[0094] Step S5, according to the Laplace manifold learning algorithm, a manifold regularization term is constructed using the label co-occurrence similarity to obtain a label co-occurrence manifold regularization term.

[0095] Among them, a sample data set with some missing labels and its corresponding label matrix are obtained from the historical sample data, and the label co-occurrence similarity is calculated using the sample data set and its corresponding label matrix.

[0096] Furthermore, combined with the Laplace manifold learning algorithm, the label co-occurrence similarity is used to construct the manifold regularization term, and the obtained label co-occurrence manifold regularization term is used as the last term in the multi-label classification model, thereby obtaining the prediction model.

[0097] The present invention uses a multi-label classification model as the basic model, takes into account the high dimensionality and redundancy of user behavior features, and uses the Hilbert Schmidt independence criterion to achieve feature extraction, thereby improving the classification performance of the model. At the same time, the label correlation is introduced by using the prior information of the user's labels, and the local correlation relationship between labels is used to better restore the customer label matrix, thereby completing the missing label information in the existing user portrait and establishing new customer labels.

[0098] In a specific embodiment of the present invention, Figure 6 The flowchart of the user tag prediction method shown in FIG. 1 specifically includes:

[0099] S101, using the commercial bank asset data platform to obtain the user behavior transaction information table, extract relevant field information from the behavior transaction data table as user features. At the same time, perform data cleaning on the feature information - remove sample data with a feature missing rate greater than 0.5.

[0100] S102, for samples with a feature missing rate less than 0.5, the user behavior feature information is restored using a sparse subspace algorithm, a restored complete feature matrix XC is obtained, and the restored input is used as input data of a preset prediction model for calculation.

[0101] S103, in order to ensure the consistency between user feature information and label information, the Hilbert-Schmidt independence index is used to maximize the feature extraction matrix P, thereby reducing the feature dimension of the sample data and improving the model classification performance. The user feature matrix after feature extraction is XCP.

[0102] S104, using the correlation between the user transaction behavior information and the user tag and the correlation between the user tag and the tag, the model algorithm of the present invention is used to obtain the feature selection matrix W. When a new user makes the current transaction, the user's own feature information XCP can be multiplied with the feature selection matrix W. The user tag information of the new user is accurately predicted.

[0103] In this embodiment, the present invention records the entire process of financial asset flow under the financial asset business of commercial banks. The new user label portrait is analyzed by using the user's debit card transaction amount, cash deposit amount, number of online banking transactions, and third-party transaction behavior data within a certain period of time as user behavior features.

[0104] Step 1: Technical analysis and preprocessing

[0105] Considering the problem of missing features in user behavior data, when the feature missing rate (i.e., the number of missing feature categories of a user / the total number of feature categories) is greater than 0.5, the information value contained in the sample data is low, so the sample is removed. When the feature missing rate is less than 0.5, the sample data sparse subspace self-representation model (SSC) is used to restore the missing features. The model is shown in formula (1). The input data in the formula is user feature data X, and the restored user complete feature data XC can be obtained by multiplying the input data X by the self-representation matrix C.

[0106] Step 2: Feature Engineering

[0107] The Hilbert-Schmidt independence index is to define the covariance operator on the reproducing kernel Hilbert space, from which a statistic suitable for measuring independence is derived to judge independence. Considering the problem of high dimensionality of user behavior feature data, the present invention starts from the user's intrinsic features, that is, the user's behavior transaction information, and uses the Hilbert-Schmidt independence index (HSIC) to maximize the dependency between user tags and features to achieve feature extraction, as shown in formula (2).

[0108] Step 3, model selection

[0109] The present invention uses the classic multi-label classification model (LSML) as the basic model. This method can predict the new sample data label after step 1 data preprocessing, i.e., missing feature recovery. The LSML algorithm model is as follows:

[0110]

[0111] Among them, X represents the user transaction behavior feature matrix, and Y represents the user known label information matrix. λ1, λ2, λ3, and λ4 are model hyperparameters used to adjust the weights. The second item of the model considers that the label information is discrete data, so the label matrix is ​​continuous using the label self-representation matrix C. The first and last items of the model respectively characterize the correlation between the feature space and the label from a global and local perspective to better obtain the feature selection matrix W and thus better predict the label information XW.

[0112] Furthermore, considering that each class label is determined by certain specific attributes of the instance itself, a multi-label high-order strategy is adopted to add a label matrix for supplementary labels to the initial label matrix, and to construct a multi-label classifier using the specific information in the classification labels and the label correlation.

[0113] Furthermore, the last term in the LSML model is based on the classic Laplace manifold learning algorithm (LE), which maps high-dimensional sample points to low-dimensional space while ensuring that the essential characteristics of the data samples remain unchanged, thereby maintaining the local structure of the data points unchanged and promoting model completion and classification capabilities. The Euclidean distance is used in the LSML model to measure label similarity, but labels are usually discrete data, so the label co-occurrence similarity measurement method is used in the present invention to reconstruct the Laplace manifold regularization term.

[0114] Specifically, given a sample data set X=[x1,x2,...,x n ]∈R n*d , label matrix Y = [Y1,Y2,...,Y l ]'∈R n*l , where d represents the number of features and l represents the number of labels. If the sample x i With the g-th label, then y ig =1, otherwise, y ig = 0. If the sample contains such label information missing, then y ig =0.5, the label co-occurrence similarity is defined as follows:

[0115]

[0116]

[0117] Obviously, if y ig With yih When both are 1, The value is 1. If y ig With y ih When both are 0, The value is the hyperparameter α. If y ig When missing, it is equal to y ih The similarity of 1 and 0 can be defined as If y ik With y il If both labels are missing, the co-occurrence can be defined as

[0118] Based on the consistency assumption of labels and the above label co-occurrence similarity, the following manifold regularization term can be constructed:

[0119]

[0120] Among them, F is the predicted user label matrix, O is the label correlation matrix, L = DO is the graph Laplacian matrix of O, and D is the degree matrix.

[0121] Furthermore, the regularization term uses the high-order correlation between labels to strengthen the correlation between labels. This improves the accuracy of the predicted known user label matrix F and the user's real label information, thereby improving the accuracy of the feature selection matrix W. When new user feature information is input, XW is used to predict the user's label information, so the more accurate W is, the more accurate the predicted user label information is.

[0122] Step 4, the present invention introduces a label co-occurrence manifold regularization term based on the LSML model to improve the prediction accuracy of the LSML model, so the proposed model algorithm is as follows:

[0123]

[0124] Among them, the model uses linear fitting method to build the connection between user feature space and user label. Hadamard product is used to ensure the consistency of predicted label and known label information. The l21 norm is used to achieve the purpose of feature selection. Finally, the label co-occurrence similarity is used to induce model parameter learning from a local perspective. When the transaction feature information X of a new user is input, the feature selection matrix W trained by the model, the feature projection matrix feature extraction matrix P and the feature self-representation matrix C obtained by data preprocessing can be used, and the user prediction label matrix can be expressed as XCPW.

[0125] The present invention improves the richness of user portraits. By completing missing tag information and establishing new customer tags, the solution can more accurately describe customer characteristics and behaviors. In this way, banks can obtain richer customer portraits and better understand user needs and behaviors. The present invention enhances business decision support. The accurate establishment of tags helps banks make more accurate business decisions, such as personalized recommendations and risk assessments. These decisions can better meet customer needs, improve user experience, and promote business development.

[0126] In addition, the present invention improves personalized services. Based on the accurate information of new user tags, banks can provide more personalized services. For different tag combinations, banks can customize different products, recommendations, marketing activities, etc. for users to meet the personalized needs of users and improve customer stickiness and loyalty. The present invention refines marketing and promotion. With more comprehensive user tag information, banks can conduct marketing and promotion activities more targeted. This can reduce resource waste, improve promotion efficiency, and deliver publicity and services to user groups that are truly interested. The establishment of new customer tags creates business expansion opportunities for banks. By deeply understanding user needs, banks can develop new products and services that are more suitable for users with different tags, thereby expanding market share and increasing business revenue.

[0127] like Figure 7 FIG. 1 is a schematic diagram of a structure of a user tag prediction device according to an embodiment of the present invention. The device shown in the figure includes:

[0128] The feature missing rate module 10 is used to obtain the behavior transaction information table authorized by the user, extract key information from the behavior transaction information table, obtain multiple user feature information, and determine the feature missing rate corresponding to each user feature information;

[0129] The data recovery module 20 is used to compare the preset missing rate threshold with the feature missing rate, perform data cleaning on the user feature information, and use a sparse subspace algorithm to recover the user feature information after data cleaning;

[0130] The label prediction module 30 is used to process the user feature information after data recovery using a pre-established prediction model to obtain a user label prediction result.

[0131] As an embodiment of the present invention, Figure 8 As shown, the feature missing rate module 10 includes:

[0132] A feature category number unit 11 is used to count the number of missing feature categories of user feature information of a single user and the total number of feature categories of user feature information of all users;

[0133] The feature missing rate unit 12 is used to determine the feature missing rate corresponding to each user feature information according to the number of missing feature categories and the total number of feature categories.

[0134] As an embodiment of the present invention, the data recovery module is further used to compare a preset missing rate threshold with the feature missing rate, and remove user feature information whose feature missing rate is greater than the preset missing rate threshold to complete data cleaning of user feature information.

[0135] As an embodiment of the present invention, Fig. 9 As shown, the data recovery module 20 includes:

[0136] A self-representation matrix unit 21, used to determine a behavior feature data matrix and a self-representation matrix according to the user feature information after data cleaning;

[0137] The data recovery unit 22 is used to recover the user characteristic information after data cleaning by using a sparse subspace algorithm according to the behavior characteristic data matrix and the self-representation matrix to obtain complete characteristic data of the user.

[0138] In this embodiment, if Fig.10 As shown, the label prediction module 30 includes:

[0139] The feature extraction matrix unit 31 is used to determine the feature extraction matrix corresponding to the complete feature data of the user according to the complete feature data of the user;

[0140] The behavior transaction matrix unit 32 is used to process the feature extraction matrix using a preset independence index calculation model to obtain a user behavior transaction matrix.

[0141] In this embodiment, the label prediction module is also used to use a pre-established multi-label classification model to perform prediction processing on the user behavior transaction matrix; wherein the multi-label classification model includes a label co-occurrence manifold regularization term.

[0142] In this embodiment, if Fig.11 As shown, the device also includes:

[0143] The co-occurrence similarity module 40 is used to obtain a sample data set with some missing labels and its corresponding label matrix, and determine the label co-occurrence similarity using the sample data set and its corresponding label matrix;

[0144] The manifold regularization term module 50 is used to construct a manifold regularization term using label co-occurrence similarity according to the Laplace manifold learning algorithm to obtain the label co-occurrence manifold regularization term.

[0145] Based on the same application concept as the above-mentioned user tag prediction method, the present invention also provides the above-mentioned user tag prediction device. Since the principle of solving the problem by the user tag prediction device is similar to that of the user tag prediction method, the implementation of the user tag prediction device can refer to the implementation of the user tag prediction method, and the repeated parts will not be repeated.

[0146] The present invention calculates the feature missing rate of user feature information, cleans and restores user feature information, effectively completes missing information, improves data accuracy, improves the efficiency of label division, saves manpower and material resources, and combines the method of predicting user labels with a pre-established prediction model. It can more accurately describe user characteristics and behaviors, better meet customer personalized needs, achieve accurate division of user labels, improve user experience, reduce resource waste, and improve promotion efficiency.

[0147] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above method is implemented when the processor executes the program.

[0148] The present invention also provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when the computer program / instruction is executed by a processor.

[0149] The present invention also provides a computer-readable storage medium, which stores a computer program for executing the above method by a computer.

[0150] like Fig.12 As shown, the electronic device 600 may further include: a communication module 110, an input unit 120, an audio processor 130, a display 160, and a power supply 170. It is worth noting that the electronic device 600 does not necessarily have to include Fig.12 In addition, the electronic device 600 may also include Fig.12 For components not shown, reference may be made to the prior art.

[0151] like Fig.12 As shown, the central processor 100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor devices and / or logic devices. The central processor 100 receives inputs and controls the operations of various components of the electronic device 600.

[0152] The memory 140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 100 may execute the program stored in the memory 140 to implement information storage or processing.

[0153] The input unit 120 provides input to the CPU 100. The input unit 120 is, for example, a key or a touch input device. The power supply 170 is used to provide power to the electronic device 600. The display 160 is used to display display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.

[0154] The memory 140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 140 may also be some other type of device. The memory 140 includes a buffer memory 141 (sometimes referred to as a buffer). The memory 140 may include an application / function storage unit 142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 600 through the central processor 100.

[0155] The memory 140 may also include a data storage unit 143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 144 of the memory 140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0156] The communication module 110 is a transmitter / receiver 110 that transmits and receives signals via an antenna 111. The communication module (transmitter / receiver) 110 is coupled to the central processor 100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0157] Based on different communication technologies, multiple communication modules 110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module. The communication module (transmitter / receiver) 110 is also coupled to a speaker 131 and a microphone 132 via an audio processor 130 to provide an audio output via the speaker 131 and receive an audio input from the microphone 132, thereby realizing a common telecommunication function. The audio processor 130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 130 is also coupled to the central processor 100, so that the sound can be recorded on the local machine through the microphone 132, and the sound stored on the local machine can be played through the speaker 131.

[0158] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0160] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0162] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A user tag prediction method, characterized in that: The method comprises: Obtaining a behavior transaction information table authorized by the user, extracting key information from the behavior transaction information table, obtaining multiple user feature information, and determining a feature missing rate corresponding to each user feature information; Comparing a preset missing rate threshold with the feature missing rate, performing data cleaning on the user feature information, and using a sparse subspace algorithm to perform data recovery on the user feature information after data cleaning; The pre-established prediction model is used to process the user feature information after data recovery to obtain the user label prediction result.

2. The method according to claim 1, characterized in that Determining the feature missing rate corresponding to each user feature information includes: Count the number of missing feature categories of user feature information of a single user, and the total number of feature categories of user feature information of all users; The feature missing rate corresponding to each user feature information is determined according to the number of missing feature categories and the total number of feature categories.

3. The method according to claim 1, characterized in that Comparing a preset missing rate threshold with the feature missing rate, and performing data cleaning on the user feature information includes: The preset missing rate threshold is compared with the feature missing rate, and the user feature information whose feature missing rate is greater than the preset missing rate threshold is eliminated to complete the data cleaning of the user feature information.

4. The method according to claim 1, characterized in that: The method of using the sparse subspace algorithm to recover the user characteristic information after data cleaning includes: According to the user characteristic information after data cleaning, determine the behavior characteristic data matrix and the self-representation matrix; According to the behavior feature data matrix and the self-representation matrix, a sparse subspace algorithm is used to recover the user feature information after data cleaning to obtain complete user feature data.

5. The method according to claim 4, characterized in that The method of processing the user characteristic information after data recovery by using the pre-established prediction model includes: Determining a feature extraction matrix corresponding to the complete feature data of the user according to the complete feature data of the user; The feature extraction matrix is ​​processed using a preset independence index calculation model to obtain a user behavior transaction matrix.

6. The method according to claim 5, characterized in that The processing of the user characteristic information after data recovery by using the pre-established prediction model also includes: The user behavior transaction matrix is ​​predicted using a pre-established multi-label classification model, wherein the multi-label classification model includes a label co-occurrence manifold regularization term.

7. The method according to claim 6, characterized in that The method further comprises: Obtaining a sample data set with some missing labels and its corresponding label matrix, and determining the label co-occurrence similarity using the sample data set and its corresponding label matrix; According to the Laplace manifold learning algorithm, the label co-occurrence similarity is used to construct a manifold regularization term to obtain the label co-occurrence manifold regularization term.

8. A user tag prediction device, characterized in that: The device comprises: A feature missing rate module is used to obtain a behavior transaction information table authorized by a user, extract key information from the behavior transaction information table, obtain multiple user feature information, and determine a feature missing rate corresponding to each user feature information; A data recovery module, used to compare a preset missing rate threshold with the feature missing rate, perform data cleaning on the user feature information, and use a sparse subspace algorithm to perform data recovery on the user feature information after data cleaning; The label prediction module is used to process the user feature information after data recovery using a pre-established prediction model to obtain a user label prediction result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for causing a computer to execute the method according to any one of claims 1 to 7.

Images