Method, apparatus, device, storage medium and program product for processing service data

Abstract

The application relates to the technical field of big data, in particular to a business data processing method and device, computer equipment, a storage medium and a computer program product. The business data to be processed is divided into multiple view data according to a preset classification dimension; for each type of view data, a feature matrix is constructed by using the view data, and a target optimization function is constructed based on the feature matrix and the view data; the target optimization function is solved to obtain a feature extraction result of the business data; and the feature extraction result is used to represent feature information of the business data. The method can improve the efficiency of obtaining the feature extraction result representing the feature information of the business data, that is, the efficiency of obtaining the feature information of the business data is improved through the process.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a method, apparatus, device, storage medium, and program product for processing business data. Background Technology

[0002] With the development of information technology, various industries have generated a large amount of business data, such as the financial industry. Effective analysis of this business data can help companies formulate more optimized operation and maintenance strategies based on the analysis results.

[0003] Traditional technologies primarily utilize machine learning to analyze business data. However, due to the large volume, high dimensionality, and complex relationships between data points in the business data being analyzed, traditional methods suffer from low efficiency in business data analysis. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, equipment, storage medium, and program product for processing business data that can improve the efficiency of business data analysis, in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for processing business data, the method comprising:

[0006] The business data to be processed is divided into multiple types of view data according to the preset classification dimensions;

[0007] For each type of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data;

[0008] The objective optimization function is solved to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0009] In one embodiment, the feature matrix is ​​a first feature matrix or a second feature matrix; the step of constructing the feature matrix using the view data for each type of view data includes:

[0010] Determine whether the business data has a tag; the tag is used to characterize the type information of the business data;

[0011] If so, the first feature matrix is ​​constructed using the view data and the corresponding labels of the view data;

[0012] If not, then the second feature matrix is ​​constructed using the view data.

[0013] In one embodiment, constructing the first feature matrix using the view data and the corresponding labels of the view data includes:

[0014] The view data is sorted to obtain sorted view data;

[0015] The first feature matrix is ​​constructed based on the labels corresponding to the sorted view data and the position information of the sorted view data.

[0016] In one embodiment, constructing the target optimization function based on the feature matrix and the view data includes:

[0017] Based on the first feature matrix and the view data, the target optimization function is constructed.

[0018] In one embodiment, the method further includes:

[0019] Obtain the target label with the largest proportion in the target view data; the target view data consists of multiple view data that are less than a preset threshold in distance from the view data.

[0020] The target label is determined as the reconstruction label of the view data, and an optimized first feature matrix is ​​constructed based on the view data and the reconstruction label corresponding to the view data;

[0021] Based on the optimized first feature matrix and the view data, an adjusted target optimization function is constructed.

[0022] The adjusted objective optimization function is solved to obtain the feature extraction results.

[0023] In one embodiment, constructing the feature matrix using the view data includes:

[0024] The second feature matrix is ​​constructed based on the nearest neighbor information of the view data.

[0025] In one embodiment, constructing the target optimization function based on the feature matrix and the view data includes:

[0026] Based on the second feature matrix and the view data, the target optimization function is constructed.

[0027] In one embodiment, solving the objective optimization function to obtain the feature extraction results of the business data includes:

[0028] Solve the objective optimization function to obtain the feature extraction matrix of the business data;

[0029] Based on the feature extraction matrix, the feature extraction result is obtained.

[0030] In one embodiment, the method further includes:

[0031] Using the feature extraction matrix and the new business data to be processed, the feature extraction results of the new business data to be processed are obtained.

[0032] Secondly, this application also provides a business data processing apparatus, the apparatus comprising:

[0033] The segmentation module is used to divide the business data to be processed into multiple types of view data according to preset classification dimensions;

[0034] The first construction module is used to construct a feature matrix using the view data for each type of view data, and to construct a target optimization function based on the feature matrix and the view data.

[0035] The first acquisition module is used to solve the target optimization function and obtain the feature extraction result of the business data; the feature extraction result is used to characterize the feature information of the business data.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0037] The business data to be processed is divided into multiple types of view data according to the preset classification dimensions;

[0038] For each type of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data;

[0039] The objective optimization function is solved to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] The business data to be processed is divided into multiple types of view data according to the preset classification dimensions;

[0042] For each type of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data;

[0043] The objective optimization function is solved to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] The business data to be processed is divided into multiple types of view data according to the preset classification dimensions;

[0046] For each type of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data;

[0047] The objective optimization function is solved to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0048] The aforementioned business data processing methods, devices, equipment, storage media, and program products can divide the business data to be processed into multiple types of view data according to a preset classification dimension. For each type of view data, a feature matrix can be constructed using the view data, and a target optimization function can be constructed based on the feature matrix and each type of view data. Since the constructed feature matrix can characterize the feature information of the view data, and the constructed feature matrix can extract low-dimensional features containing useful information of the data in the high-dimensional data feature space, the dimension of the obtained feature matrix is ​​lower than that of the view data. This allows for the rapid construction of the target optimization function based on the feature matrix and the view data, improving the efficiency of constructing the target optimization function. The feature extraction result of the business data is obtained by solving the target optimization function, which also improves the efficiency of obtaining the feature extraction result that characterizes the feature information of the business data. In other words, this process improves the efficiency of obtaining the feature information of the business data. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for processing business data in one embodiment;

[0050] Figure 2 This is a flowchart illustrating a business data processing method in another embodiment;

[0051] Figure 3 This is a flowchart illustrating a business data processing method in another embodiment;

[0052] Figure 4 This is a flowchart illustrating a business data processing method in another embodiment;

[0053] Figure 5This is a flowchart illustrating a method for processing tagged business data in one embodiment;

[0054] Figure 6 This is a flowchart illustrating a method for processing business data without tags in one embodiment;

[0055] Figure 7 This is a structural block diagram of a business data processing device in one embodiment;

[0056] Figure 8 This is a structural block diagram of a business data processing apparatus in another embodiment;

[0057] Figure 9 This is a structural block diagram of a business data processing apparatus in another embodiment;

[0058] Figure 10 This is a structural block diagram of a business data processing apparatus in another embodiment;

[0059] Figure 11 This is a structural block diagram of a business data processing apparatus in another embodiment;

[0060] Figure 12 This is a structural block diagram of a business data processing apparatus in another embodiment;

[0061] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0063] It should be noted that the business data processing methods, apparatus, equipment, storage media, and program products of this application can be applied in the field of big data, as well as in other technical fields. This application does not limit the application fields of the business data processing methods, apparatus, equipment, storage media, and program products.

[0064] In one embodiment, such as Figure 1 As shown, a method for processing business data is provided. This embodiment illustrates the application of this method to a computer device. It is understood that this method can also be applied to a server, and further to a system including both a computer device and a server, and is implemented through the interaction between the computer device and the server. In this embodiment, the method includes the following steps:

[0065] S101, divide the business data to be processed into multiple types of view data according to the preset classification dimensions.

[0066] The business data to be processed can be financial business data, enterprise approval business data, substation business data, etc.; the preset classification dimensions can be the type of business data, the user group corresponding to the business data, etc. For example, taking financial business data as the business data to be processed in this embodiment, the preset classification dimensions can be three categories: user basic information, user asset information, and user behavior information. It is understood that in this embodiment, the categories of view data correspond to the classification dimensions. For example, if the classification dimensions include four dimensions, the resulting view data will be of four categories; or, for example, if the classification dimensions include five dimensions, the resulting view data will be of five categories. Optionally, the preset classification dimensions can be determined based on historical experience values ​​or according to the data type contained in the business data to be processed.

[0067] S102: For various types of view data, construct a feature matrix using the view data, and construct a target optimization function based on the feature matrix and view data.

[0068] Optionally, in this embodiment, a feature matrix can be constructed for each type of view data using the nearest neighbor relationship between the view data and adjacent view data. Alternatively, a feature matrix can be constructed using the labels corresponding to the view data. Or, a preset feature extraction method can be used to extract features from the view data and construct a feature matrix using the extracted features.

[0069] Optionally, in this embodiment, the objective optimization function can be composed of principal component terms, regression terms, and regularization terms. The principal component terms can be constructed using the aforementioned feature matrix, the regression terms can be constructed based on the view data, and the regression terms can be fitted with a nonlinear mapping using linear projection, enabling the feature extraction method to have nonlinear dimensionality reduction capabilities and effectively improving feature extraction performance. The regularization term can also be constructed based on the view data to alleviate the overfitting problem of the objective optimization function and improve its robustness. For example, the constructed objective optimization function can be:

[0070]

[0071]

[0072] In the formula, B represents the principal component term constructed based on the feature matrix; v is the number of views; and γ is a hyperparameter used to adjust the influence of the regularization term on the entire optimization problem. For regression terms, a nonlinear mapping can be fitted using linear projection to construct the feature extraction method, enabling nonlinear dimensionality reduction and effectively improving the feature extraction performance. As a regularization term, it can alleviate overfitting and improve model robustness; To maximize the divergence of variables after feature extraction, the correlation between variables can be reduced, which can effectively improve the accuracy of subsequent model learning; α i Y, P (i) Let α be the objective function of the optimization problem. i P is the adaptive view weight balancing factor. (i) Y represents the feature extraction dimensionality reduction matrix for each view; Y is the feature extraction result for the current data.

[0073] S103, Solve the objective optimization function to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0074] Optionally, in this embodiment, based on the optimization conditions corresponding to the above-mentioned target optimization function, the hyperparameters and initialization parameters of the target optimization function can be set, the target optimization function can be solved, the feature extraction matrix corresponding to the above-mentioned business data can be obtained, and the feature extraction result representing the feature information of the above-mentioned business data can be obtained based on the feature extraction matrix.

[0075] In the above-mentioned business data processing method, the business data to be processed can be divided into multiple types of view data according to the preset classification dimensions. For each type of view data, a feature matrix can be constructed using the view data, and a target optimization function can be constructed based on the feature matrix and each type of view data. Since the constructed feature matrix can characterize the feature information of the view data, and the constructed feature matrix can extract low-dimensional features containing useful information of the data in the high-dimensional data feature space, the dimension of the obtained feature matrix is ​​lower than that of the view data. In this way, the target optimization function can be quickly constructed based on the feature matrix and the view data, which improves the efficiency of constructing the target optimization function. The feature extraction result of the business data is obtained by solving the target optimization function, which also improves the efficiency of obtaining the feature extraction result that characterizes the feature information of the business data. In other words, this process improves the efficiency of obtaining the feature information of the business data.

[0076] In some scenarios, the aforementioned business data can be tagged business data or untagged business data. Different methods can be used to construct the aforementioned feature matrix for different types of business data. In the scenario described above, where feature matrices are constructed using various types of view data, the constructed feature matrix can be a first feature matrix or a second feature matrix. In one embodiment, such as... Figure 2As shown, "Constructing a feature matrix using view data for various types of view data" in S102 includes:

[0077] S201, determine whether the business data has a label; the label is used to characterize the type information of the business data.

[0078] The tags on the business data can be used to characterize the type information of the business data. Optionally, the aforementioned business data can be tagged business data or untagged business data. Optionally, in this embodiment, the computer device can determine whether the business data has a tag based on whether the business data has corresponding type information. For example, if the business data is of type K, that is, the business data has corresponding type information, then the computer device can determine that the business data is tagged business data.

[0079] S202, if so, then construct the first feature matrix using the view data and the corresponding labels of the view data.

[0080] In this embodiment, for business data with tags, the computer device can construct a first feature matrix using view data and the tags corresponding to the view data. Optionally, in one embodiment, the view data can be sorted first to obtain sorted view data. The first feature matrix is ​​then constructed based on the tags corresponding to the sorted view data and the position information of the sorted view data. For example, the process of sorting the view data can be to classify and sort the view data according to the tag information of the view data. Accordingly, each category of view data after classification and sorting will have a corresponding feature matrix. Optionally, the first feature matrix constructed in this embodiment can be an indicator matrix. For example, using... Let the indicator matrix represent the k-th class. Then, the constructed indicator matrix is ​​as follows:

[0081]

[0082] S203, if not, then construct the second feature matrix using the view data.

[0083] In this embodiment, for cases where business data lacks labels, the computer device can construct a second feature matrix using view data. Optionally, in one embodiment, the second feature matrix can be constructed based on the nearest neighbor information of the view data, where the nearest neighbor information can be the nearest neighbor relationships between view data across all views. Optionally, the second feature matrix constructed in this embodiment can be a global-local structure preservation matrix W. For example, the global-local structure preservation matrix W constructed using the nearest neighbor relationships of view data can be:

[0084] If vi and v j Nearest neighbor in all views

[0085] It should be noted that the global local structure preservation matrix W constructed in this embodiment simultaneously considers the nearest neighbor relationship between view data under all views and the distance between global nearest neighbor view data under different views.

[0086] In this embodiment, different feature matrices are constructed by determining whether the business data has labels. If the business data has labels, a first feature matrix is ​​constructed using the view data and the corresponding labels. If the business data does not have labels, a second feature matrix is ​​constructed based on the nearest neighbor information of the view data. Since this embodiment constructs corresponding feature matrices for different types of view data, by classifying the view data, it can fully utilize the label information and multi-view information of the labeled business data. This preserves the core characteristics of each view data while compensating for their deficiencies, making the dimensionality-reduced business data more comprehensive and accurate. When constructing the second feature matrix for business data without labels, the nearest neighbor relationship between view data under all views and the distance between global nearest neighbor view data under different views are considered simultaneously. This is of great significance for maintaining the global topological relationship after data dimensionality reduction.

[0087] Furthermore, in the scenario of constructing a target optimization function based on the feature matrix and view data, if the constructed feature matrix is ​​the first feature matrix, then in one embodiment, "constructing a target optimization function based on the feature matrix and view data" in S102 includes: constructing a target optimization function based on the first feature matrix and view data.

[0088] Optionally, the constructed target optimization function can be a supervised optimization function built based on the first feature matrix and multi-view data. As an optional implementation, in this embodiment, the target optimization function can be a supervised optimization function, composed of an information discrimination term, a regression term, and a regularization term, and the view weights are adjusted through an adaptive method. For example, in this embodiment, the target optimization function constructed based on the first feature matrix and view data can be:

[0089]

[0090]

[0091] In the formula, For information discrimination terms constructed based on the first feature matrix, the feature extraction effect can be improved by aggregating similar view data as closely as possible, α i Y, P (i) Let α be the objective function of the optimization problem. iP is the adaptive view weight balancing factor. (i) Y represents the feature extraction dimensionality reduction matrix for each view; Y is the feature extraction result for the current data.

[0092] In this embodiment, for business data with labels, the objective optimization function constructed based on the first feature matrix and the view data can improve the feature extraction effect by aggregating similar sample points as closely as possible, which can make the dimensionality-reduced business data more comprehensive and the constructed objective optimization function more accurate.

[0093] In scenarios where the business data consists of tagged data, the labels may be inaccurate. Therefore, the labels of the business data can be reconstructed, and an optimized target function can be constructed using the business data and the reconstructed labels. In one embodiment, such as... Figure 3 As shown, the above method also includes:

[0094] S301, Obtain the target label with the largest proportion in the target view data; the target view data consists of multiple view data that are less than a preset threshold distance from the view data.

[0095] The target view data can be multiple view data that are less than a preset threshold in distance from the view data. It is understood that the preset threshold can be set in advance based on empirical values. Optionally, in this embodiment, the computer device can use the K-nearest neighbor method to determine multiple target view data that are less than the preset threshold in distance from the view data. Further, for the target label with the largest proportion in the target view data, it can be obtained by statistically analyzing all label data in the target view to obtain the target label with the largest proportion. It should be noted that if multiple labels in the target view data have the same proportion, the label corresponding to the target view data that is closest to the view data in the target view data corresponding to these labels can be selected as the target label.

[0096] S302, the target label is determined as the reconstruction label of the view data, and the optimized first feature matrix is ​​constructed based on the view data and the corresponding reconstruction label of the view data.

[0097] In this embodiment, the computer device determines the target label obtained above as the reconstruction label of the view data. Further, the computer device can use the method for constructing the first feature matrix in S202 above to construct an optimized first feature matrix based on the view data and the corresponding reconstruction label. It should be noted that the process of constructing the optimized first feature matrix is ​​the same as the process described in S202 above, and will not be repeated here.

[0098] S303, construct the adjusted objective optimization function based on the optimized first feature matrix and view data.

[0099] For example, in this embodiment, with Taking the optimized first feature matrix as an example, the adjusted objective function constructed based on the optimized first feature matrix and the view data can be:

[0100]

[0101]

[0102] In the formula, This represents the adjusted objective function. This represents the optimized first feature matrix. For an explanation of the other parameters in the formula, please refer to the description in S202 above. This embodiment will not repeat them here.

[0103] S304, solve the adjusted optimization function to obtain the feature extraction results.

[0104] In this embodiment, the process of obtaining the feature extraction results can be to first set the initial value of the hyperparameter γ and initialize the parameter α. i =1 / v, then solve the above objective function and calculate the intermediate data. Set the loop termination condition: A≤σ or the loop count reaches k times, and use the following formula to iteratively calculate the result:

[0105] (1) Solve using the generalized eigenvalue method The solution to Y is a matrix composed of the eigenvectors corresponding to the first M smallest eigenvalues.

[0106] (2) Solve

[0107]

[0108] Wherein, α is obtained i For view weights, further, based on the above solution results, the feature extraction matrix is ​​obtained. Where Y represents the feature extraction result.

[0109] In this embodiment, the target label with the largest proportion among multiple view data that are less than a preset threshold away from the view data is determined as the reconstruction label of the view data. An optimized first feature matrix is ​​constructed based on the view data and the corresponding reconstruction label. An adjusted target optimization function is constructed based on the optimized first feature matrix and the view data. The adjusted target optimization function is solved to obtain the feature extraction result. This feature extraction method, while aggregating similar view data and separating dissimilar view data with the help of label information, also corrects the label data. It can extract differences between highly related features, remove invalid and redundant information, improve the quality of business data, reduce the dimensionality of business data, reduce the scale of business data storage, increase the value of business data, and reduce the management cost of business data. It can effectively solve various problems caused by excessively high business data dimensionality and inaccurate business data labels, and can effectively improve the accuracy and efficiency of subsequent business data processing, avoiding overfitting.

[0110] Furthermore, in the scenario of constructing a target optimization function based on the feature matrix and view data, if the constructed feature matrix is ​​the second feature matrix, then in one embodiment, "constructing a target optimization function based on the feature matrix and view data" in S102 includes: constructing a target optimization function based on the second feature matrix and view data.

[0111] The objective optimization function can be an unsupervised optimization function constructed based on the second feature matrix and multi-view data. As an optional implementation, in this embodiment, the objective optimization function can be an unsupervised optimization function, composed of a global-local structure preservation term, a regression term, and a regularization term, and the view weights are adjusted using an adaptive method. For example, in this embodiment, the objective optimization function constructed based on the second feature matrix and view data can be:

[0112]

[0113]

[0114] In the formula, The global-local structure preservation term is constructed based on the second feature matrix, which can measure the local nearest neighbor relationship between view data from the perspective of all view data; v is the number of views, and γ is a hyperparameter used to adjust the impact of the regularization term on the entire optimization problem; For regression terms, a nonlinear mapping can be fitted using linear projection to construct the feature extraction method, enabling nonlinear dimensionality reduction and effectively improving the feature extraction performance. As a regularization term, it can alleviate overfitting and improve model robustness; To maximize the divergence of variables after feature extraction, the correlation between variables can be reduced, which can effectively improve the accuracy of subsequent model learning; α i Y, P (i) Let α be the objective function of the optimization problem. i P is the adaptive view weight balancing factor. (i) Y represents the feature extraction dimensionality reduction matrix for each view; Y is the feature extraction result for the current data.

[0115] In this embodiment, for business data without labels, since the objective optimization function constructed based on the second feature matrix can measure the local nearest neighbor relationship between view data from the perspective of all view data, it can make full use of information from different view data, retaining the core characteristics of each view data and making up for missing information, thus achieving the effect of making the dimensionality-reduced business data more comprehensive and accurate.

[0116] In the scenario described above, where the objective optimization function is solved to obtain the feature extraction results of the business data, the objective optimization function can be solved to obtain the feature extraction matrix of the business data, and the feature extraction results of the business data can be obtained based on the feature extraction matrix. In one embodiment, such as... Figure 4 As shown, S103 above includes:

[0117] S401, solve the objective optimization function to obtain the feature extraction matrix of the business data.

[0118] In this embodiment, the process of solving the above objective optimization function can be as follows: set the initial value of the hyperparameter γ, and initialize the parameter α. i =1 / v, calculate the intermediate data to obtain the feature extraction matrix of the business data. It's understandable that the objective optimization function constructed for labeled and unlabeled business data is different; therefore, the formulas for solving the intermediate data are also different. These two cases will be explained separately below. Optionally, for labeled business data, the formula... Calculate its intermediate data; for business data without tags, the formula for calculating the intermediate data can be: In the formula, L is the Laplace matrix based on the global-local structure preservation matrix; furthermore, the loop termination condition can be set: A≤σ or the number of loop iterations reaches k, and the result can be calculated iteratively using the following formula:

[0119] (1) Solve using the generalized eigenvalue method The solution to Y is a matrix composed of the eigenvectors corresponding to the first M smallest eigenvalues.

[0120] (2) Solve

[0121]

[0122] In the formula, α i Furthermore, based on the above solution results, the feature extraction matrix can be obtained as the view weights.

[0123] S402, based on the feature extraction matrix, obtain the feature extraction results.

[0124] In this embodiment, the computer device obtains the feature extraction matrix of the business data through the above-described solution process. In the formula, Y represents the feature extraction result corresponding to the business data. After obtaining the feature extraction matrix, the computer equipment can extract Y from the feature extraction matrix to obtain the feature extraction result of the business data.

[0125] In this embodiment, by solving the constructed objective optimization function, the feature extraction matrix of the business data can be accurately obtained. Thus, based on the feature extraction matrix, the feature extraction results of the business data can be accurately obtained, improving the accuracy of the obtained feature extraction results.

[0126] After obtaining the aforementioned feature extraction matrix, if new business data needs to be processed, the obtained feature extraction matrix can be directly used to process the new business data. This process will be described in detail below. In one embodiment, the above method further includes: using the feature extraction matrix and the new business data to be processed to obtain the feature extraction result of the new business data to be processed.

[0127] Specifically, in this embodiment, the new business data to be processed can be tagged business data or untagged business data. It is understood that if the new business data to be processed is tagged business data, the computer device can calculate... Quickly obtain the corresponding feature extraction results, where X (i) This indicates new business data to be processed. This represents the feature extraction matrix corresponding to labeled business data; if the new business data to be processed is unlabeled business data, the computer equipment can calculate... Quickly obtain the corresponding feature extraction results, where X (i) This indicates new business data to be processed. This represents the feature extraction matrix corresponding to unlabeled business data.

[0128] In this embodiment, for new business data to be processed, the computer device can quickly obtain the feature extraction results of the new business data to be processed by using the already acquired feature extraction matrix and the new business data to be processed, thereby improving the efficiency of obtaining the feature extraction results of the new business data to be processed.

[0129] For example, when the business data to be processed is business data with tags, please refer to [link to relevant documentation]. Figure 5 The processing procedure for this business data can be seen in the following example: Taking 500-dimensional banking business data as an example, firstly, the 500-dimensional business data can be split into 5 view data according to customer basic information, transaction information, asset information, product holding information, and behavioral information, and then an indicator matrix C is constructed. k Given the objective function A, with hyperparameter γ = 0.5, and the loop termination condition A ≤ 0.001 or the number of loops reaching 50, solve for the objective function and obtain M. (i) The low-dimensional representation and view weights of the view data are iteratively solved after H until the loop termination condition is met. The resulting Y is the initial feature extraction result. For the initial feature extraction matrix, α i Let Y be the view weight. Further, labels can be reconstructed for each view data point in the initial feature extraction matrix, and an optimized indicator matrix can be constructed based on the original view data labels and the reconstructed labels. An objective optimization function can then be constructed based on the optimized indicator matrix. Hyperparameters and iteration stopping conditions are set to solve the objective optimization function, yielding a new feature extraction result Y. and α i Furthermore, any newly added business data awaiting processing can be calculated. Quickly obtain the feature extraction results of newly added business data to be processed.

[0130] For example, when the business data to be processed is business data without tags, please refer to [link to relevant documentation]. Figure 6 The processing procedure for this business data can be seen in the following example: Taking 500-dimensional banking business data as an example, firstly, the 500-dimensional data can be split into five view data according to customer basic information, transaction information, asset information, product holding information, and behavioral information. Then, a global-local structure preservation matrix W and an objective optimization function are constructed. The hyperparameter γ = 0.5 and the loop termination condition A ≤ 0.001 or the number of loops reaches 50 are set. The objective optimization function is solved, and M is calculated. (i) The low-dimensional representation and view weights of the view data are iteratively solved after H until the loop termination condition is met. The resulting Y is the initial feature extraction result. For the initial feature extraction matrix, αi This is the view weight. If new business data is added and needs to be processed, it can be calculated... Quickly obtain the feature extraction results of newly added business data to be processed.

[0131] To facilitate understanding by those skilled in the art, the following provides a detailed description of the processing method for tagged business data, which may include:

[0132] S1: For the business data to be processed, analyze, break down, and classify it based on the real-world significance of each dimension's features to form multi-view data.

[0133] S2 constructs an indicator matrix based on all view data and labels. The indicator matrix representing the k-th class is constructed as follows:

[0134]

[0135] S3. For the generated multi-view data, a supervised objective optimization function is constructed. The objective optimization function consists of an information discrimination term, a regression term, and a regularization term, and the view weights are adjusted using an adaptive method. The specific objective optimization function is as follows:

[0136]

[0137]

[0138] In the formula, v is the number of views, and γ is a hyperparameter used to adjust the impact of the regularization term on the overall optimization problem; For information discrimination, feature extraction performance can be improved by aggregating similar sample points as closely as possible. For regression terms, a nonlinear mapping can be fitted using linear projection to construct the feature extraction method, enabling nonlinear dimensionality reduction and effectively improving the feature extraction performance. As a regularization term, it can alleviate overfitting and improve model robustness; To maximize the divergence of variables after feature extraction, the correlation between variables can be reduced, which can effectively improve the accuracy of subsequent model learning; α i Y, P (i) Let α be the objective function of the optimization problem. i P is the adaptive view weight balancing factor. (i) Y represents the feature extraction dimensionality reduction matrix for each view; Y is the feature extraction result for the current data.

[0139] S4, Set the hyperparameter γ to a suitable value, and initialize the parameter α. i =1 / v.

[0140] S5, Solve the objective function and calculate intermediate data.

[0141]

[0142] The objective function is optimized to extract the results, and the loop termination condition is set as follows: A ≤ σ or the number of iterations reaches k. The results are then calculated iteratively using the following formula:

[0143] (2) Solve

[0144]

[0145] S6, further, based on the above solution results, the feature extraction matrix is ​​obtained. In the formula, Y represents the feature extraction result corresponding to the business data.

[0146] S7. Reconstruct the labels for the target view data by finding the K nearest neighbor view data. The reconstruction label for the target view data is the label with the largest proportion among the K nearest neighbor view data. If multiple labels have the same proportion in the target view data, the label corresponding to the nearest neighbor view data among the view data corresponding to these labels can be selected as the reconstruction label for the target view data.

[0147] S8, constructs an optimized indicator matrix based on all original view data and reconstructed labels. Repeat steps S3 to S8 to obtain the data-corrected feature matrix. and feature extraction results

[0148] S9, for new business data to be processed, can be calculated. Quickly obtain the corresponding feature extraction results.

[0149] In addition, for business data without tags, the following provides a detailed description of the processing methods for business data provided in this disclosure, which may include:

[0150] S1: For the business data to be processed, analyze, break down, and classify it based on the real-world significance of each dimension's features to form multi-view data.

[0151] S2, Construct a global-local structure preservation matrix based on all view data:

[0152] If v i and v j They are nearest neighbors in all views.

[0153] S3. For the generated multi-view data, an unsupervised objective optimization function is constructed. The objective optimization function consists of a global-local structure preservation term, a regression term, and a regularization term, and the view weights are adjusted using an adaptive method. The specific objective optimization function is as follows:

[0154]

[0155]

[0156] In the formula, v is the number of views, and γ is a hyperparameter used to adjust the impact of the regularization term on the overall optimization problem; As a global-local structure preservation item, it can measure the local proximity relationship between view data from the perspective of the entire view data; For regression terms, a nonlinear mapping can be fitted using linear projection to construct the feature extraction method, enabling nonlinear dimensionality reduction and effectively improving the feature extraction performance. As a regularization term, it can alleviate overfitting and improve model robustness; To maximize the divergence of variables after feature extraction, the correlation between variables can be reduced, which can effectively improve the accuracy of subsequent model learning; α i Y, P (i) Let α be the objective function of the optimization problem. i P is the adaptive view weight balancing factor. (i) Y represents the feature extraction dimensionality reduction matrix for each view; Y is the feature extraction result for the current data.

[0157] S4, Set the hyperparameter γ to a suitable value, and initialize the parameter α. i =1 / v.

[0158] S5, Solve the objective function and calculate intermediate data.

[0159] L is the Laplacian matrix based on the global-local structure-preserving matrix; the objective function is extracted, and the loop termination condition is set: A ≤ σ or the number of iterations reaches k. The results are calculated iteratively using the following formula: the generalized eigenvalue method is used to solve the problem. The solution to Y is a matrix composed of the eigenvectors corresponding to the first M smallest eigenvalues. Solving for Y...

[0160] S6, Based on the above solution results, the feature extraction matrix is ​​obtained. In the formula, Y represents the feature extraction result.

[0161] S7, for new data to be processed, calculations can be performed. Quickly obtain the corresponding feature extraction results.

[0162] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0163] Based on the same inventive concept, this application also provides a business data processing apparatus for implementing the above-described business data processing method. The solution provided by this apparatus is similar to the implementation described in the above-described method; therefore, the specific limitations in one or more business data processing apparatus embodiments provided below can be found in the limitations of the business data processing method described above, and will not be repeated here.

[0164] In one embodiment, such as Figure 7 As shown, a business data processing apparatus is provided, comprising: a partitioning module 10, a first construction module 11, and a first acquisition module 12, wherein:

[0165] The segmentation module 10 is used to divide the business data to be processed into multiple types of view data according to preset classification dimensions.

[0166] The first construction module 11 is used to construct a feature matrix using the view data for various types of view data, and to construct a target optimization function based on the feature matrix and the view data.

[0167] The first acquisition module 12 is used to solve the objective optimization function and obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0168] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0169] In one embodiment, the feature matrix is ​​either a first feature matrix or a second feature matrix, such as... Figure 8 As shown, the first construction module 11 includes: a determining unit 111, a first construction unit 112, and a second construction unit 113, wherein:

[0170] The determining unit 111 is used to determine whether the business data has a label; the label is used to characterize the type information of the business data.

[0171] The first construction unit 112 is used to construct a first feature matrix using the view data and the corresponding labels of the view data if the business data has labels.

[0172] The second construction unit 113 is used to construct a second feature matrix using view data if the business data does not have labels.

[0173] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0174] In one embodiment, the first construction unit 112 is used to sort the view data to obtain sorted view data; and to construct a first feature matrix based on the labels corresponding to the sorted view data and the position information of the sorted view data.

[0175] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0176] In one embodiment, such as Figure 9 As shown, the first building module 11 further includes a third building unit 114, wherein:

[0177] The third building unit 114 is used to build a target optimization function based on the first feature matrix and view data.

[0178] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0179] In one embodiment, such as Figure 10 As shown, the above-mentioned device further includes: a second acquisition module 13, a reconstruction module 14, a second construction module 15, and a third acquisition module 16, wherein:

[0180] The second acquisition module 13 is used to determine the target label as the reconstruction label of the view data, and to construct an optimized first feature matrix based on the view data and the reconstruction label corresponding to the view data.

[0181] The reconstruction module 14 is used to determine the target label as the reconstruction label of the view data, and to construct the optimized first feature matrix based on the view data and the reconstruction label corresponding to the view data.

[0182] The second building module 15 is used to build an adjusted target optimization function based on the optimized first feature matrix and view data.

[0183] The third acquisition module 16 is used to solve the adjusted objective optimization function and obtain the feature extraction results.

[0184] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0185] In one embodiment, the second construction unit 113 is used to construct a second feature matrix based on the nearest neighbor information of the view data.

[0186] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0187] In one embodiment, the third building unit 114 is used to build a target optimization function based on the second feature matrix and view data.

[0188] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0189] In one embodiment, such as Figure 11 As shown, the first acquisition module 12 includes: a solution unit 121 and an acquisition unit 122, wherein:

[0190] Solver 121 is used to solve the objective optimization function and obtain the feature extraction matrix of the business data.

[0191] The acquisition unit 122 is used to acquire feature extraction results based on the feature extraction matrix.

[0192] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0193] In one embodiment, such as Figure 12 As shown, the above-mentioned device further includes: a fourth acquisition module 17, wherein:

[0194] The fourth acquisition module 17 is used to acquire the feature extraction results of the new business data to be processed by utilizing the feature extraction matrix and the new business data to be processed.

[0195] The business data processing device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0196] Each module in the aforementioned business data processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0197] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 13 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores business data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for processing business data.

[0198] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0199] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0200] The business data to be processed is divided into multiple view data according to the preset classification dimensions.

[0201] For various types of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data.

[0202] Solve the objective optimization function to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0203] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0204] The business data to be processed is divided into multiple view data according to the preset classification dimensions.

[0205] For various types of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data.

[0206] Solve the objective optimization function to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0207] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0208] The business data to be processed is divided into multiple view data according to the preset classification dimensions.

[0209] For various types of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data.

[0210] Solve the objective optimization function to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data.

[0211] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0212] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0213] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0214] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for processing business data, characterized in that, The method includes: The business data to be processed is divided into multiple types of view data according to the preset classification dimensions; For each type of view data, a feature matrix is ​​constructed using the view data, and a target optimization function is constructed based on the feature matrix and the view data; The objective optimization function is solved to obtain the feature extraction results of the business data; the feature extraction results are used to characterize the feature information of the business data. The feature matrix is ​​either a first feature matrix or a second feature matrix, wherein the first feature matrix is ​​constructed using the view data and the labels corresponding to the view data; When the feature matrix is ​​a first feature matrix, the step of constructing a target optimization function based on the feature matrix and the view data includes: Based on the first feature matrix and the view data, the target optimization function is constructed; The method further includes: Obtain the target label with the largest proportion in the target view data; the target view data consists of multiple view data that are less than a preset threshold in distance from the view data. The target label is determined as the reconstruction label of the view data, and an optimized first feature matrix is ​​constructed based on the view data and the reconstruction label corresponding to the view data; Based on the optimized first feature matrix and the view data, an adjusted target optimization function is constructed. The adjusted objective optimization function is solved to obtain the feature extraction results.

2. The method according to claim 1, characterized in that, The step of constructing a feature matrix using the view data for each type of view data includes: Determine whether the business data has a tag; the tag is used to characterize the type information of the business data; If so, the first feature matrix is ​​constructed using the view data and the corresponding labels of the view data; If not, then the second feature matrix is ​​constructed using the view data.

3. The method according to claim 2, characterized in that, The step of constructing the first feature matrix using the view data and the corresponding labels includes: The view data is sorted to obtain sorted view data; The first feature matrix is ​​constructed based on the labels corresponding to the sorted view data and the position information of the sorted view data.

4. The method according to claim 2, characterized in that, The process of constructing the feature matrix using the view data includes: The second feature matrix is ​​constructed based on the nearest neighbor information of the view data.

5. The method according to claim 4, characterized in that, The construction of the target optimization function based on the feature matrix and the view data includes: Based on the second feature matrix and the view data, the target optimization function is constructed.

6. The method according to claim 1, characterized in that, Solving the objective optimization function to obtain the feature extraction results of the business data includes: Solve the objective optimization function to obtain the feature extraction matrix of the business data; Based on the feature extraction matrix, the feature extraction result is obtained.

7. The method according to claim 6, characterized in that, The method further includes: Using the feature extraction matrix and the new business data to be processed, the feature extraction results of the new business data to be processed are obtained.

8. A business data processing apparatus, characterized in that, The device includes: The segmentation module is used to divide the business data to be processed into multiple types of view data according to preset classification dimensions; The first construction module is used to construct a feature matrix using the view data for each type of view data, and to construct a target optimization function based on the feature matrix and the view data. The first acquisition module is used to solve the target optimization function and acquire the feature extraction result of the business data; the feature extraction result is used to characterize the feature information of the business data. The feature matrix is ​​either a first feature matrix or a second feature matrix, wherein the first feature matrix is ​​constructed using the view data and the labels corresponding to the view data; When the feature matrix is ​​a first feature matrix, the first construction module is specifically used for: Based on the first feature matrix and the view data, the target optimization function is constructed; The device further includes: a second acquisition module, a reconstruction module, a second construction module, and a third acquisition module. The second acquisition module is used to acquire the target label with the largest proportion in the target view data; the target view data consists of multiple view data that are less than a preset threshold distance from the view data. The reconstruction module is used to determine the target label as the reconstruction label of the view data, and to construct an optimized first feature matrix based on the view data and the reconstruction label corresponding to the view data. The second construction module is used to construct an adjusted target optimization function based on the optimized first feature matrix and the view data; The third acquisition module is used to solve the adjusted target optimization function and obtain the feature extraction result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images