Data processing method and device, computer, storage medium and program product

Abstract

The embodiment of the invention discloses a data processing method and device, a computer, a storage medium and a program product, and relates to a data transmission technology in the field of big data, and the method comprises the steps: obtaining initial feature data of first sample data corresponding to A data categories; performing singular value decomposition on the A initial feature data to obtain A first orthogonal matrixes, and determining a corresponding principal component matrix based on the A first orthogonal matrixes; acquiring first feature data of the second sample data, projecting the first feature data by adopting the principal component matrixes corresponding to the A data categories respectively to obtain category spatial features corresponding to the second sample data in the A data categories respectively, and performing feature fusion to obtain sample category features; inputting the sample category features into the initial classification model for model training to obtain a target classification model; the target classification model is used for performing classification prediction in A data categories. According to the invention, the accuracy of classification prediction can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a data processing method, apparatus, computer, storage medium, and program product. Background Art

[0002] Multi-class classification problems have wide applications in various fields, such as image classification, natural language processing, medical diagnosis, etc. Multi-class classification problems can also be applied to object recognition and detection tasks. For example, in the field of autonomous driving, objects on the road surface are classified into categories such as cars, pedestrians, traffic signs, etc. for intelligent assisted driving and traffic safety monitoring. Multi-class classification problems in these fields all require the establishment of appropriate data sets and the application of machine learning algorithms, such as logistic regression, support vector machines, decision trees, neural networks, random forests, K-nearest neighbors, etc., to train models and perform prediction and classification. However, in practical applications, logistic regression and support vector machines based on multi-class classification problems have low computational efficiency. Especially when the number of classes is large, a high computational cost is required; the training of random forests and neural networks also requires a large amount of time and computational resources, and continuous improvement of algorithms and more powerful computational capabilities are needed to make the application scope of multi-class classification problems wider and provide possibilities for solving many practical problems. Summary of the Invention

[0003] Embodiments of this application provide a data processing method, apparatus, computer, storage medium, and program product, which can improve the computational efficiency of model training and classification prediction while reducing the dimension of feature vectors.

[0004] On the one hand, embodiments of this application provide a data processing method, which includes:

[0005] Obtain first sample data corresponding to A data classes respectively, and perform feature extraction on the first sample data corresponding to A data classes respectively to obtain A initial feature data; A is a positive integer;

[0006] Perform singular value decomposition on A initial feature data respectively to obtain A first orthogonal matrices, and determine principal component matrices corresponding to A data classes respectively based on A first orthogonal matrices;

[0007] Obtain first feature data of second sample data, project the first feature data of second sample data by using principal component matrices corresponding to A data classes respectively to obtain class space features corresponding to second sample data in A data classes respectively, and fuse the A class space features to obtain sample class features;

[0008] Input the sample category features into the initial classification model for prediction to obtain the sample prediction results corresponding to the second sample data. Adjust the parameters of the initial classification model based on the sample prediction results and the data category labels of the second sample data to obtain the target classification model. The target classification model is used for classification prediction among A data categories.

[0009] In one aspect, an embodiment of the present application provides a data processing device, which includes:

[0010] A data acquisition module, configured to acquire first sample data corresponding to A data categories respectively;

[0011] A feature extraction module, configured to extract features from the first sample data corresponding to A data categories respectively to obtain A initial feature data; A is a positive integer;

[0012] A data decomposition module, configured to perform singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and determine the principal component matrices corresponding to the A data categories based on the A first orthogonal matrices;

[0013] A data processing module, configured to acquire the first feature data of the second sample data, project the first feature data of the second sample data by using the principal component matrices corresponding to the A data categories respectively to obtain the category space features corresponding to the second sample data in the A data categories respectively, and fuse the A category space features to obtain the sample category features;

[0014] A model training module, configured to input the sample category features into the initial classification model for prediction to obtain the sample prediction results corresponding to the second sample data, and adjust the parameters of the initial classification model based on the sample prediction results and the data category labels of the second sample data to obtain the target classification model. The target classification model is used for classification prediction among A data categories.

[0015] In a possible implementation manner, the data acquisition module is further configured to perform the following operations:

[0016] Acquire the sample data to be classified, send the sample data to be classified to M service devices for classification processing, and receive the initial category labels of the sample data to be classified returned by each service device;

[0017] Integrate and process the initial category labels of the sample data to be classified by each of the M service devices respectively to obtain the classification category labels of the sample data to be classified;

[0018] Based on the classification category labels of the sample data to be classified, split the sample data to be classified into the first sample data corresponding to the A data categories respectively.

[0019] In a possible implementation, the data decomposition module is used to perform singular value decomposition on A initial feature data respectively to obtain A first orthogonal matrices. When determining the principal component matrices corresponding to A data categories based on the A first orthogonal matrices, the data decomposition module specifically performs the following operations:

[0020] Perform centering offset processing on the initial feature data corresponding to the t-th data category to obtain a first sample matrix, and perform transpose processing on the first sample matrix to obtain a first transposed matrix; t is a positive integer less than or equal to A;

[0021] Fuse the features of the first sample matrix and the first transposed matrix to obtain the first orthogonal matrix and the second orthogonal matrix corresponding to the t-th data category;

[0022] Obtain the eigenvalue data of the first orthogonal matrix, decompose the eigenvalue data to obtain the singular value matrix corresponding to the t-th data category;

[0023] According to the singular value matrix, obtain the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category. When t is equal to A, obtain the principal component matrices corresponding to A data categories respectively.

[0024] In a possible implementation, when the data decomposition module is used to obtain the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category according to the singular value matrix, the data decomposition module specifically performs the following operations:

[0025] Arrange the singular values in the singular value matrix in descending order to obtain a first sequence;

[0026] Obtain a limiting coefficient, determine a first singular value based on the singular values in the singular value matrix and the limiting coefficient, and determine a first limiting parameter based on the first sequence and the first singular value; the first limiting parameter is used to make the first singular value located between a first constraint data and a second constraint data; the first constraint data is obtained by processing the first sequence with the first limiting parameter, the second constraint data is obtained by processing the first sequence with a second limiting parameter, and the second limiting parameter is updated from the first limiting parameter;

[0027] Based on the first limiting parameter, determine L column data in the first orthogonal matrix corresponding to the t-th data category as the principal component matrix corresponding to the t-th data category; L is a positive integer and L is the first limiting parameter.

[0028] In a possible implementation, the data processing module is used to project the first feature data of the second sample data respectively by using the principal component matrices corresponding to A data categories, so as to obtain the category space features corresponding to the second sample data in A data categories respectively. When the A category space features are feature - spliced and fused to obtain the sample category features, the data processing module is specifically used to perform the following operations:

[0029] In the t - th data category, perform a central offset on the first feature data of the second sample data to obtain the offset feature of the second sample data in the t - th data category;

[0030] Use the principal component matrix corresponding to the t - th data category to perform a spatial mapping on the offset feature of the second sample data in the t - th data category, so as to obtain the category space feature of the second sample data in the t - th data category;

[0031] When the category space features corresponding to the second sample data in A data categories are obtained respectively, perform feature fusion on the category space features corresponding to the second sample data in A data categories respectively to obtain the sample category feature corresponding to the second sample data.

[0032] In a possible implementation, the sample prediction result includes the category probabilities for A data categories of the second sample data. When the model training module is used to adjust the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain the target classification model, it is specifically used to perform the following operations:

[0033] Obtain the data category label corresponding to the second sample data, and generate a loss function according to the data category label and A category probabilities;

[0034] Use the loss function to adjust the parameters of the initial classification model until the parameters converge to obtain the target classification model.

[0035] In a possible implementation, the data processing device further includes a classification and prediction module, and the classification and prediction module is specifically used to perform the following operations:

[0036] Obtain the data to be predicted, extract features from the data to be predicted to obtain the second feature data, project the second feature data by using the principal component matrices corresponding to A data categories respectively to obtain the to - be - predicted space features corresponding to the data to be predicted in A data categories respectively, and perform feature fusion on the A to - be - predicted space features to obtain the to - be - predicted category feature;

[0037] Input the feature of the category to be predicted into the target classification model for prediction, and obtain the prediction probabilities corresponding to the data to be predicted in A data categories respectively. Determine the data category with the maximum prediction probability as the classification prediction result of the data to be predicted.

[0038] One aspect of the embodiments of the present application provides a computer device, including a processor, a memory, and an input / output interface;

[0039] The processor is respectively connected to the memory and the input / output interface. Among them, the input / output interface is used to receive and output data, the memory is used to store computer programs, and the processor is used to call the computer programs so that the computer device including the processor executes the method in one aspect of the embodiments of the present application.

[0040] One aspect of the embodiments of the present application provides a computer-readable storage medium storing a computer program, which is adapted to be loaded and executed by a processor so that a computer device having the processor executes the method in one aspect of the embodiments of the present application.

[0041] One aspect of the embodiments of the present application provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device executes the methods provided in various alternative manners in one aspect of the embodiments of the present application. In other words, when the computer instructions are executed by the processor, the methods provided in various alternative manners in one aspect of the embodiments of the present application are implemented.

[0042] Implementing the embodiments of the present application will have the following beneficial effects:

[0043] In the embodiments of the present application, first sample data corresponding to A data categories are obtained, and feature extraction is performed on the first sample data corresponding to the A data categories respectively to obtain A initial feature data; A is a positive integer; singular value decomposition is respectively performed on the A initial feature data to obtain A first orthogonal matrices, and based on the A first orthogonal matrices, principal component matrices corresponding to the A data categories are determined; the first feature data of the second sample data is obtained, and the first feature data of the second sample data is projected by using the principal component matrices corresponding to the A data categories respectively to obtain category space features corresponding to the second sample data in the A data categories respectively, and the A category space features are feature fused to obtain sample category features; the sample category features are input into an initial classification model for prediction to obtain a sample prediction result corresponding to the second sample data, and the initial classification model is parameter adjusted through the sample prediction result and the data category label of the second sample data to obtain a target classification model; the target classification model is used for classification prediction among the A data categories. Through the above process, the methods of feature extraction and model optimization are fully utilized to map the features of the sample data into the category spaces of different data categories, so that the features of the sample data can better show the correlation with each data category, effectively improving the accuracy of the classification model, and thus better coping with the classification problem in the case of complex data. By methods such as singular value decomposition and principal component analysis, key features are extracted to reduce the dimension of the data, effectively reducing the redundant information and noise interference of the data, and improving the robustness of the classification model and the calculation efficiency of classification prediction. At the same time, the strategy of parameter adjustment by combining the sample prediction result output by the model and the data type label makes the classification model more conform to the actual data distribution, and improves the accuracy and stability of the classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 It is a network interaction architecture diagram for data processing provided by an embodiment of the present application;

[0046] Figure 2 It is a schematic diagram of a data processing scenario provided by an embodiment of the present application;

[0047] Figure 3 It is a method flow of data processing provided by an embodiment of the present application Figure 1 ;

[0048] Figure 4 is a method flow for data processing provided by an embodiment of the present application Figure 2 ;

[0049] Figure 5 is a schematic diagram of a data processing device provided by an embodiment of the present application;

[0050] Figure 6 is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0051] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0052] Among them, if it is necessary to collect data of an object (such as a user, etc.) in the present application, a prompt interface or a pop-up window is displayed before and during the collection. The prompt interface or the pop-up window is used to prompt the user that some data is being collected currently. Only after obtaining the confirmation operation of the user on the prompt interface or the pop-up window, the relevant steps for data acquisition are started, otherwise it ends. Moreover, the obtained user data will be used in reasonable, legal scenarios or uses, etc. Optionally, in some scenarios where user data needs to be used but the user's authorization has not been obtained, authorization can also be requested from the user, and the user data will be used when the authorization is passed.

[0053] Among them, the present application may involve machine learning technology in the field of artificial intelligence, and the training and use of the model are realized through machine learning technology.

[0054] Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results of theory, methods, technologies, and application systems. In other words, artificial intelligence is a comprehensive technology of computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence is also to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning, and decision-making. For example, study the process of text annotation and text parsing, and generate a method that can parse the text annotation and text parsing results in a way similar to human intelligence.

[0055] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, autonomous driving, and intelligent transportation.

[0056] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning. For example, in this application, the training and use of the modal feature parsing model and data parsing model corresponding to each data modality, etc., are carried out by training the model so that the model continuously learns new knowledge or skills, and then a trained model is obtained for data parsing. For example, this application is to learn the technology for data parsing to obtain a trained modal feature parsing modality and data parsing model, etc., so that the modal feature parsing modality and data parsing model can be used to parse interactive data.

[0057] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in multiple fields. For example, common ones include smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless, autonomous driving, drones, robots, intelligent healthcare, intelligent customer service, vehicle networking, autonomous driving, and intelligent transportation. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0058] In the embodiments of this application, please refer to Figure 1 , Figure 1 is a network interaction architecture diagram for data processing provided by the embodiments of this application. As Figure 1 shown, the computer device 101 can obtain the first sample data corresponding to A data categories from the service device, or can obtain the first sample data corresponding to A data categories from the memory space of the computer device 101, or can simultaneously obtain the first sample data corresponding to A data categories from both the service device and the memory space of the computer device 101, etc. Among them, the number of service devices is one or more, so thatFigure 1 For example, the service devices may include service device 102a, service device 102b, service device 102c, etc. Among them, data interaction can be carried out between the service devices, or the service devices can carry out data interaction through the computer device 101, etc. Optionally, the computer device 101 can also obtain public data from the Internet, classify it, and obtain the first sample data corresponding to A data categories respectively, etc., which will not be limited here. Among them, when the computer device 101 obtains the first sample data corresponding to A data categories respectively from the service device, there is the first sample data corresponding to A data categories respectively in the service device, such as the first sample data 1021 corresponding to A data categories respectively in the service device 102a. Among them, the computer device 101 can obtain the first sample data corresponding to A data categories respectively, extract features from the first sample data corresponding to A data categories respectively, obtain A initial feature data, determine the principal component matrix corresponding to A data categories respectively based on the A initial feature data, project the first feature data of the second sample data by using the above A principal component matrices, and obtain the category space features corresponding to the second sample data in A data categories respectively, so as to realize the dimensionality reduction processing of the sample data, effectively reduce the redundant information and noise interference of the data, and improve the calculation efficiency of classification prediction; fuse the A category space features; perform model training based on the A category space features, and adjust the parameters through the sample prediction results and the data category labels of the second sample data to obtain the target classification model, so that the target classification model is more in line with the actual data distribution and improves the accuracy and stability of classification prediction.

[0059] Specifically, please refer to Figure 2 , Figure 2 which is a schematic diagram of a data processing scenario provided by an embodiment of the present application. As Figure 2 shown, the computer device can obtain the first sample data corresponding to A data categories respectively, process the first sample data corresponding to A data categories respectively, and obtain the principal component matrix 201 corresponding to A data categories respectively. A is a positive integer, such as Figure 2 the principal component matrix 1, the principal component matrix 2, and the principal component matrix A shown. Obtain the first feature data of the second sample data, project the first feature data of the second sample data by using the principal component matrix corresponding to A data categories respectively, and obtain A category space features 202, such as Figure 2 the category space feature 1, the category space feature 2, and the category space feature A shown. Fuse the A category space features to obtain the sample category feature, input the sample category feature into the initial classification model 203 for prediction, and obtain the category probability 204 of the second sample data in A data categories, such as Figure 2The category probabilities such as category probability 1, category probability 2, and category probability A shown above determine the classification result 205 of the second sample data based on the numerical magnitudes of the A category probabilities. For example, if category probability 1 is 0.8, category probability 2 is 0, category probability A is 0.1, etc., and it should be noted that the sum of the numerical values of the A category probabilities is 1, then the first data category corresponding to category probability 1 is determined as the classification result of the second sample data. Through, for example, Figure 2 the sample prediction result and the data category label of the second sample data shown above, the parameters of the initial classification model 203 are adjusted until the parameters of the adjusted initial classification model converge, and the initial classification model when the parameters converge is determined as the target classification model. Through the above process, the sample category features have all the features of the corresponding sample data in each data category space, making the training effect of the initial classification model better, thereby improving the accuracy and stability of classification prediction.

[0060] It can be understood that the service device mentioned in the embodiments of the present application can also be a computer device. The computer device in the embodiments of the present application includes, but is not limited to, a terminal device or a server. In other words, the computer device can be a server or a terminal device, or a system composed of a server and a terminal device. Among them, the terminal device mentioned above can be an electronic device, including but not limited to a mobile phone, a tablet computer, a desktop computer, a laptop computer, a handheld computer, a vehicle-mounted device, an augmented reality / virtual reality (AR / VR) device, a head-mounted display, a smart TV, a wearable device, a smart speaker, a digital camera, a camera, and other mobile internet devices (MIDs) with network access capabilities, or terminal devices in scenarios such as trains, ships, and flights. As Figure 1 shown in, the terminal device can be a laptop computer (as shown by service device 102b), a mobile phone (as shown by service device 102c), or a vehicle-mounted device (as shown by service device 102a), etc. Figure 1 Only some of the devices are listed as examples. Optionally, the service device 102a refers to a device located in the vehicle 103. Among them, the server mentioned above can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, vehicle-road coordination, content delivery network (CDN), and big data and artificial intelligence platforms.

[0061] Optionally, the data involved in the embodiments of the present application may be stored in a computer device, or the data may be stored based on cloud storage technology or a blockchain network, which is not limited herein.

[0062] Further, please refer to Figure 3 , Figure 3 which is a method flow for data processing provided by the embodiments of the present application Figure 1 . As Figure 3 shown, it is used to describe the model training process, and the data processing process includes the following steps:

[0063] Step S101: Obtain first sample data corresponding to A data categories respectively, and perform feature extraction on the first sample data corresponding to A data categories respectively to obtain A initial feature data; A is a positive integer.

[0064] In the embodiments of the present application, a computer device may obtain sample data to be classified, send the sample data to be classified to M service devices for classification processing, and receive the initial category labels of the sample data to be classified returned by each service device; integrate the initial category labels of the sample data to be classified by each of the M service devices respectively to obtain the classification category labels of the sample data to be classified; based on the classification category labels of the sample data to be classified, split the sample data to be classified into first sample data corresponding to A data categories respectively. Optionally, the computer device may obtain a data set from the public network, and access a website or service providing data by using a public API (Application Programming Interface) interface to obtain the required data; in addition, some websites also provide data download services, and the required data set can be directly downloaded. The above data set should contain public data categories, and the data corresponding to A data categories respectively are extracted from the data set as the first sample data. It should be noted that when obtaining data on the network, it is necessary to ensure the legality and usage rights of the data, and attention should be paid to the quality and source of the data to ensure the reliability and effectiveness of the data.

[0065] Specifically, after obtaining the first sample data, it is also necessary to transform the first sample data into feature data with higher representational ability for subsequent model training and analysis. Through processes such as feature selection, feature transformation, feature construction, and feature extraction, the first sample data corresponding to A data categories will be subjected to feature extraction to obtain A initial feature data. Among them, feature selection is to select the most relevant and useful features according to the requirements of the classification problem and the characteristics of the data, and statistical methods (such as correlation analysis) or domain knowledge can be used for feature selection; feature transformation is to appropriately transform the data to meet the requirements of the model. Common feature transformations include standardization or normalization of numerical features, encoding of discrete features (such as one-hot encoding), etc.; feature construction is to construct new features according to the background of the classification problem and the characteristics of the data to improve the expressive ability of the model, which can include combination, differentiation, correlation, etc. of features; feature extraction is to select appropriate feature extraction methods according to the type of data and the nature of the features. Common feature extraction methods include statistical features (such as mean, variance), frequency domain features (such as Fourier transform coefficients), time series features (such as time window statistics), etc.

[0066] Specifically, the first sample data corresponding to the t-th data category can be obtained, and the first sample features of the first sample data corresponding to the t-th data category can be obtained. Among them, the number of the first sample data corresponding to the t-th data category is denoted as nt, and nt is a positive integer. The first sample features of the nt first sample data corresponding to the t-th data category can be feature-stitched to obtain the initial feature data corresponding to the t-th data category, where t is a positive integer less than or equal to A. Similarly, the initial feature data corresponding to A data categories can be obtained. For example, for the initial feature data corresponding to the t-th data category, the first sample feature of each first sample data is represented by x i (x i can be an m-dimensional vertical feature vector), where m is a positive integer. Then the initial feature data corresponding to the t-th data category can be represented as X t = [x1, x2, …, x nt .

[0067] Step S102: Perform singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and based on the A first orthogonal matrices, determine the principal component matrices corresponding to the A data categories respectively.

[0068] In an embodiment of the present application, the computer device performs singular value decomposition on A initial feature data to obtain A first orthogonal matrices. Taking the t-th data category as an example, the computer device performs a central offset process on the initial feature data corresponding to the t-th data category to obtain a first sample matrix, and performs a transpose process on the first sample matrix to obtain a first transpose matrix; t is a positive integer less than or equal to A; the first sample matrix and the first transpose matrix are subjected to feature fusion to obtain the first orthogonal matrix and the second orthogonal matrix corresponding to the t-th data category; for example, for the initial feature data corresponding to the t-th data category, the initial feature data of each sample data is represented by x i (where (x i can be an m-dimensional vertical feature vector), then the initial feature data corresponding to the t-th data category can be represented as X t =[x1, x2, …, x nt , and the central offset process is performed on the initial feature data through the central data feature, where the central data feature refers to the statistical value of the initial feature data corresponding to the t-th data category, such as the mean or median, etc. The first sample matrix after the central offset process can be represented as The first sample matrix After performing the transpose process, the first transpose matrix The first sample matrix The first transpose matrix Feature fusion (such as matrix multiplication, etc.) is performed to obtain the first orthogonal matrix corresponding to the t-th data category and the second orthogonal matrix It can be understood that the first orthogonal matrix is an m*m matrix, and the second orthogonal matrix is an n*n matrix. Among them, is used to represent the central data feature corresponding to the t-th data category. In one possible way, the central data feature can be obtained by referring to the method shown in Formula ①:

[0069]

[0070] As shown in Formula ①, is used to represent the mean value of the initial feature data corresponding to the t-th data category, nt represents that the number of the first sample features included in the initial feature data corresponding to the t-th data category is nt, and the value of i ranges from 1 to nt. For example, when i = 1, x1 represents the first sample feature in the initial feature data corresponding to the t-th data category.

[0071] Optionally, when the computational amount of performing matrix multiplication on the first sample matrix and the first transpose matrix is large, the computer device can use a more efficient and accurate iterative solution method to avoid Calculation. Among them, the computer device can construct a random matrix, select a small random matrix G, whose dimension size is usually n*k (k is much smaller than n), and then multiply the first sample matrix by the random matrix G to obtain a matrix Y of m*k, that is Then perform matrix truncation. By performing orthogonal triangular decomposition or other dimensionality reduction methods on the matrix Y, a matrix Q of m*p (p is much smaller than m) is obtained, and Q satisfies the orthogonal property; perform SVD decomposition on the matrix to obtain an approximate first orthogonal matrix U′ t , and linearly combine the approximate first orthogonal matrix U′ t with the matrix Q to obtain the first orthogonal matrix U y .

[0072] The computer device obtains the eigenvalue data of the first orthogonal matrix, decomposes the eigenvalue data, and obtains the singular value matrix corresponding to the t-th data category; optionally, the singular value matrix can also be obtained from the eigenvalue data of the second orthogonal matrix. Among them, the process of obtaining the eigenvalue data of the first orthogonal matrix can be: based on the first orthogonal matrix and the identity matrix E, determine the characteristic matrix Perform data processing on the characteristic matrix to obtain the characteristic determinant Expand the above characteristic determinant into an equation, solve the equation, and obtain m values of λ, where λ is the eigenvalue data of the first orthogonal matrix, and the identity matrix E is a matrix with elements on the diagonal from the upper left corner to the lower right corner (i.e., the main diagonal) all being 1; perform square root processing on each eigenvalue data to obtain m singular values (i.e., the singular values corresponding to the t-th data category, which can be denoted as σ), and combine the m σ into the singular value matrix corresponding to the t-th data category. Among them, the singular value matrix is a diagonal matrix, that is, elements other than the main diagonal are all 0, and the singular value matrix D t corresponding to the t-th data category can be expressed as diag(σ1,σ2,...,σ m ).

[0073] The computer device obtains the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category according to the singular value matrix. When t is equal to A, the principal component matrices corresponding to A data categories are obtained.

[0074] Specifically, the computer device arranges the singular values in the singular value matrix in descending order to obtain a first sequence; obtains a restriction coefficient, determines a first singular value based on the singular values in the singular value matrix and the restriction coefficient, and determines a first restriction parameter based on the first sequence and the first singular value; the first restriction parameter is used to make the first singular value lie between a first constraint data and a second constraint data; the first constraint data is obtained by processing the first sequence with the first restriction parameter, the second constraint data is obtained by processing the first sequence with a second restriction parameter, and the second restriction parameter is obtained by updating the first restriction parameter; based on the first restriction parameter, L columns of data in the first orthogonal matrix corresponding to the t-th data category are determined as the principal component matrix corresponding to the t-th data category; L is a positive integer and L is the first restriction parameter.

[0075] Singular Value Decomposition (SVD) is a commonly used matrix decomposition method. When the computer device operates on the singular value matrix, it can arrange the singular values in the singular value matrix in descending order to obtain a first sequence; obtains a restriction coefficient α. The restriction coefficient α is usually related to the singular values of the singular value matrix and can be determined according to specific applications and requirements; this restriction coefficient can be a constant between 0 and 1, denoted as 0 < a < 1. This restriction coefficient is used to make the finally determined first restriction parameter able to represent the key features in the t-th data category. Therefore, a constant close to 1 can be taken. For example, it can be 0.8 or 0.9, etc. Based on the singular values in the singular value matrix and the restriction coefficient, a first singular value is determined. The first singular value is obtained by calculating the product of the sum of the squares of m singular values and the restriction coefficient. For example, the first singular value can be expressed as The first singular value is used to measure the importance of the singular values. A first restriction parameter L is determined based on the first sequence and the first singular value. The first restriction parameter is usually obtained by performing operations on the first sequence and the first singular value. The first singular value is made to lie between the first constraint data and the second constraint data through the first restriction parameter. This process can be achieved by adjusting the value of the first restriction parameter. Among them, the first constraint data can be expressed as The second constraint data can be expressed as After determining the value of the first restriction parameter L, the computer device can, based on the first restriction parameter, determine L columns of data in the first orthogonal matrix corresponding to the t-th data category as the principal component matrix corresponding to the t-th data category. The principal component matrix is composed of extracting specific columns from the first orthogonal matrix and is used to represent the features or categories of the data. Among them, the first restriction parameter L determines the dimension of the principal component matrix. By adjusting the value of L, principal component matrices of different dimensions can be obtained to meet the requirements of specific tasks.

[0076] Among them, the limiting coefficient α is approximately 1, which is used to indicate the sum of the squares of m singular values in the approximate singular value matrix among the above first singular values; the first singular value The first constraint data The second constraint data The relationship between them can be expressed as And

[0077] Specifically, this process can be recorded as: obtaining the j-th first detection parameter "L j ", converting the j-th limiting parameter to be detected into the j-th second detection parameter "L j -1", where j is a positive integer; obtaining the statistical value (such as the sum of squares, etc.) of the singular values in the singular value matrix corresponding to the t-th data category to obtain singular data, and performing weighted processing on the singular data using the limiting coefficient to obtain the first singular value. Based on the j-th first detection parameter, L j singular values are obtained from the singular values in the singular value matrix corresponding to the t-th data category, and the statistical value of the L j singular values is determined as the j-th first detection constraint, and L j is the j-th first detection parameter; L j -1 singular values are obtained from the singular values in the singular value matrix corresponding to the t-th data category, and the statistical value of the L j -1 singular values is determined as the j-th second detection constraint, and L j -1 is the j-th second detection parameter. If the first singular value is between the j-th first detection constraint and the j-th second detection constraint, then the j-th first detection parameter is determined as the first limiting parameter. At this time, the j-th first detection constraint is the first constraint data, and the j-th second detection constraint is the second constraint data. If the first singular value is not between the j-th first detection constraint and the j-th second detection constraint, then the j-th first detection parameter is adjusted to obtain the (j + 1)-th first detection parameter, and the (j + 1)-th first detection parameter is detected until the first limiting parameter is obtained. Optionally, when j = 1, the j-th first detection parameter is 1, then the adjustment of the j-th first detection parameter can be an incremental process, such as the (j + 1)-th first detection parameter = the j-th first detection parameter + 1; when j = 1, the j-th first detection parameter is m, where m is a positive integer and m is the number of singular values included in the singular value matrix corresponding to the t-th data category, then the adjustment of the j-th first detection parameter can be a decremental process, such as the (j + 1)-th first detection parameter = the j-th first detection parameter - 1, etc.

[0078] Step S103: Obtain the first feature data of the second sample data. Use the principal component matrices corresponding to A data categories respectively to project the first feature data of the second sample data, obtain the category space features corresponding to the second sample data in the A data categories respectively, and perform feature fusion on the A category space features to obtain the sample category feature.

[0079] In the embodiments of the present application, the computer device obtains the first feature data of the second sample data. The above-mentioned second sample data may be a part of the sample data in the first sample data or all of the sample data, or a new sample data with the same data format as the first sample data, which is not limited herein. In the t-th data category, perform central offset on the first feature data of the second sample data to obtain the offset feature of the second sample data in the t-th data category; use the processing method of subtracting the mean value from the first feature data to perform central offset on the first feature data of the second sample data, and the above mean value is the mean value of the sample data in the t-th data category. Use the principal component matrix corresponding to the t-th data category to perform spatial mapping on the offset feature of the second sample data in the t-th data category to obtain the category space feature of the second sample data in the t-th data category; that is, the computer device uses the principal component matrix of the t-th data category to project the second sample data. Taking a sample data x k as an example, the category space feature x kt in the t-th data category can be expressed as where is the principal component matrix corresponding to the t-th data category, the superscript T in is used to represent transpose, is the mean value of the first sample data in the t-th data category. When the category space features corresponding to the second sample data in the A data categories are obtained respectively, perform feature fusion on the category space features corresponding to the second sample data in the A data categories respectively to obtain the sample category feature corresponding to the second sample data. When the computer device obtains the category space features corresponding to the second sample data in the A data categories respectively, by performing feature fusion on the category space features corresponding to the A data categories respectively, combine them into a feature vector, and this feature vector is the sample category feature corresponding to the second sample data. Among them, this sample category feature includes all the features of the second sample data in the A data category spaces. For example, the sample data x k in the second sample data, the category space features corresponding to it in the A data categories can be expressed as The sample data x k in the second sample data The corresponding sample category feature

[0080] Step S104: Input the sample class features into the initial classification model for prediction to obtain the sample prediction result corresponding to the second sample data. Adjust the parameters of the initial classification model based on the sample prediction result and the data class label of the second sample data to obtain the target classification model. The target classification model is used for classification prediction among A data classes.

[0081] In the embodiment of the present application, the computer device takes the sample class features corresponding to the second sample data obtained in the above process as input parameters and inputs them into the initial classification model that has not been trained. Since the initial classification model has not been trained, its parameters are randomly initialized. Therefore, the initial classification model calculates the input features with these randomly initialized parameters to obtain the sample prediction result. During this process, the initial classification model adjusts the parameters of the initial classification model based on the sample prediction result and the data class label of the second sample data to obtain the target classification model. A possible calculation formula for the target classification model can be seen in Formula ②:

[0082] p = softmax(Wx + b) ②

[0083] As shown in Formula ②, p is used to represent the output result obtained by predicting the input parameter x using this classification model, that is, the class probability of the input parameter x in A data classes. W and b are learnable model parameters, and the initial classification model obtains the final target classification model by training to change these two model parameters. Among them, W is a matrix with dimensions of A * m, b is a vector with dimensions of A, and the output result p is a vector with dimensions of A, and each value in it corresponds to the probability that the input parameter is this data class.

[0084] Specifically, the computer device obtains the data class label corresponding to the second sample data, generates a loss function based on the data class label and A class probabilities; uses the loss function to adjust the parameters of the initial classification model until the parameters converge to obtain the target classification model. A possible loss function can be seen in Formula ③:

[0085]

[0086] As shown in Formula ③, y it is used to represent the label corresponding to the second sample data x i for the t-th data class. If x i belongs to the t-th data class, then y it = 1. If x i does not belong to the t-th data class, then y it = 0. log is used to represent logarithmic transformation, and p it is used to represent the second sample data x iIn the output result p after passing through the initial classification model, the value at the t-th position, which is the second sample data x i The category probability belonging to the t-th data category, p it The larger it is, the greater the possibility that the corresponding sample data belongs to the t-th data category. When adjusting the parameters of the initial classification model, it is necessary to make the value of formula ③ as small as possible.

[0087] Furthermore, please refer to Figure 4 , Figure 4 which is a method flow of data processing provided by an embodiment of the present application Figure 2 . As Figure 4 shown, it is used to describe the model prediction process. The data processing process includes the following steps:

[0088] Step S201, obtain the data to be predicted, extract features from the data to be predicted to obtain second feature data, project the second feature data using the principal component matrices respectively corresponding to A data categories to obtain the to-be-predicted space features respectively corresponding to the data to be predicted in the A data categories, and fuse the A to-be-predicted space features to obtain the to-be-predicted category features.

[0089] In the embodiment of the present application, the process of the computer device extracting features from the data to be predicted can refer to Figure 4 the process of generating the initial feature data therein. After obtaining the second feature data, in the t-th data category, perform a central offset on the second feature data to obtain the to-be-predicted offset feature of the data to be predicted in the t-th data category; adopt the processing method of subtracting the mean value from the second feature data to perform a central offset on the second feature data of the data to be predicted, and the above mean value is the mean value of the sample data in the t-th data category. Use the principal component matrix corresponding to the t-th data category to perform a spatial mapping on the to-be-predicted offset feature of the data to be predicted in the t-th data category to obtain the to-be-predicted space feature of the second sample data in the t-th data category, and continue to project the second feature data using the principal component matrices of the remaining data categories until the to-be-predicted space features of the data to be predicted in the A data categories are obtained, and fuse the A to-be-predicted space features to obtain the to-be-predicted category features corresponding to the data to be predicted.

[0090] Step S202, input the to-be-predicted category features into the target classification model for prediction to obtain the prediction probabilities respectively corresponding to the data to be predicted in the A data categories, and determine the classification prediction result of the data to be predicted as the data category with the largest prediction probability.

[0091] In the embodiments of the present application, the target classification model is obtained by training an initial classification model with the sample category features of the second data sample. The computer device uses the obtained category features to be predicted as input parameters and inputs them into the target classification model for prediction. The output result is the prediction probabilities corresponding to the data to be predicted in the A data categories respectively. The computer device compares the values of the prediction probabilities and determines the data category corresponding to the prediction probability with the largest value as the classification prediction result of the data to be predicted. For example, at this time, A is equal to 3, and the three data categories are (cat, chicken, dog). The data to be predicted is an animal picture at this time. After feature extraction of the animal picture to obtain the feature data corresponding to the animal picture, the principal component matrices of the above three data categories are used to project the feature data corresponding to the animal picture to obtain the category space features corresponding to the animal picture. The three category space features are spliced and fused to obtain the category features corresponding to the animal picture. The category features corresponding to the animal picture are input into the target classification model trained with data in the same format as the animal picture for prediction. The classification probabilities obtained are (0.1, 0.9, 0), so the final classification prediction result indicates that the animal in the animal picture is a chicken.

[0092] Further, please refer to Figure 5 , Figure 5 which is a schematic diagram of a data processing device provided by an embodiment of the present application. The data processing device can be a computer program (including program code, etc.) running on a computer device. For example, the device can be an application software. The device can be used to execute the corresponding steps in the method provided by the embodiments of the present application. As Figure 5 shown, the device 500 can be used for Figure 3 and Figure 4 the computer devices corresponding to the corresponding embodiments. Specifically, the device can include: a data acquisition module 11, a feature extraction module 12, a data decomposition module 13, a data processing module 14, a model training module 15, and a classification prediction module 16.

[0093] The data acquisition module 11 is used to acquire the first sample data corresponding to each of the A data categories;

[0094] The feature extraction module 12 is used to extract features from the first sample data corresponding to each of the A data categories to obtain A initial feature data; A is a positive integer;

[0095] The data decomposition module 13 is used to perform singular value decomposition on each of the A initial feature data to obtain A first orthogonal matrices, and based on the A first orthogonal matrices, determine the principal component matrices corresponding to each of the A data categories;

[0096] The data processing module 14 is used to obtain the first feature data of the second sample data, project the first feature data of the second sample data by using the principal component matrices respectively corresponding to A data categories, obtain the category space features respectively corresponding to the second sample data in the A data categories, and perform feature fusion on the A category space features to obtain the sample category features;

[0097] The model training module 15 is used to input the sample category features into the initial classification model for prediction to obtain the sample prediction result corresponding to the second sample data, and adjust the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain the target classification model; the target classification model is used for classification prediction among the A data categories.

[0098] In a possible implementation manner, the data acquisition module 11 is further used to perform the following operations:

[0099] Obtain the sample data to be classified, send the sample data to be classified to M service devices for classification processing, and receive the initial category labels of the sample data to be classified returned by each service device;

[0100] Integrate and process the initial category labels of the sample data to be classified by each of the M service devices respectively to obtain the classification category label of the sample data to be classified;

[0101] Based on the classification category label of the sample data to be classified, split the sample data to be classified into the first sample data respectively corresponding to the A data categories.

[0102] In a possible implementation manner, when the data decomposition module 13 is used to perform singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and determine the principal component matrices respectively corresponding to the A data categories based on the A first orthogonal matrices, the data decomposition module is specifically used to perform the following operations:

[0103] Perform central offset processing on the initial feature data corresponding to the t-th data category to obtain the first sample matrix, and perform transpose processing on the first sample matrix to obtain the first transposed matrix; t is a positive integer less than or equal to A;

[0104] Perform feature fusion on the first sample matrix and the first transposed matrix to obtain the first orthogonal matrix and the second orthogonal matrix corresponding to the t-th data category;

[0105] Obtain the eigenvalue data of the first orthogonal matrix, and perform decomposition on the eigenvalue data to obtain the singular value matrix corresponding to the t-th data category;

[0106] According to the singular value matrix, obtain the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category. When t is equal to A, obtain the principal component matrices corresponding to the A data categories respectively.

[0107] In a possible implementation, when the data decomposition module 13 is used to obtain the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category according to the singular value matrix, the data decomposition module is specifically used to perform the following operations:

[0108] Arrange the singular values in the singular value matrix in descending order to obtain the first sequence;

[0109] Obtain a constraint coefficient, determine the first singular value based on the singular values in the singular value matrix and the constraint coefficient, and determine the first constraint parameter based on the first sequence and the first singular value; the first constraint parameter is used to make the first singular value lie between the first constraint data and the second constraint data; the first constraint data is obtained by processing the first sequence with the first constraint parameter, the second constraint data is obtained by processing the first sequence with the second constraint parameter, and the second constraint parameter is updated from the first constraint parameter;

[0110] Based on the first constraint parameter, determine L columns of data in the first orthogonal matrix corresponding to the t-th data category as the principal component matrix corresponding to the t-th data category; L is a positive integer and L is the first constraint parameter.

[0111] In a possible implementation, when the data processing module 14 is used to project the first feature data of the second sample data respectively by using the principal component matrices corresponding to the A data categories to obtain the category space features corresponding to the second sample data in the A data categories respectively, and perform feature splicing and fusion on the A category space feature first projection vectors to obtain the sample category feature, the data processing module is specifically used to perform the following operations:

[0112] In the t-th data category, perform central offset on the first feature data of the second sample data to obtain the offset feature of the second sample data in the t-th data category;

[0113] Use the principal component matrix corresponding to the t-th data category to perform spatial mapping on the offset feature of the second sample data in the t-th data category to obtain the category space feature of the second sample data in the t-th data category;

[0114] When obtaining the category space features corresponding to the second sample data in the A data categories respectively, perform feature fusion on the category space features corresponding to the second sample data in the A data categories respectively to obtain the sample category feature corresponding to the second sample data.

[0115] In a possible implementation, the sample prediction result includes the category probabilities of A data categories for the second sample data; when the model training module 15 is used to adjust the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain the target classification model, it is specifically used to perform the following operations:

[0116] Obtain the data category label corresponding to the second sample data, and generate a loss function according to the data category label and the A category probabilities;

[0117] Adopt the loss function to adjust the parameters of the initial classification model until the parameters converge to obtain the target classification model.

[0118] In a possible implementation, the data processing device further includes a classification prediction module 16, and the classification prediction module is specifically used to perform the following operations:

[0119] Obtain the data to be predicted, extract features from the data to be predicted to obtain second feature data, project the second feature data using the principal component matrices corresponding to the A data categories respectively to obtain the to-be-predicted space features corresponding to the data to be predicted in the A data categories respectively, and fuse the A to-be-predicted space features to obtain the to-be-predicted category features;

[0120] Input the to-be-predicted category features into the target classification model for prediction to obtain the prediction probabilities corresponding to the data to be predicted in the A data categories respectively, and determine the data category with the largest prediction probability as the classification prediction result of the data to be predicted.

[0121] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 6 shown, the computer device in the embodiment of the present application may include: a processor 601, a network interface 604, and a memory 605. In addition, the above computer device 600 may further include: a user interface 603, and at least one communication bus 602. Among them, the communication bus 602 is used to realize the connection and communication between these components. Among them, the user interface 603 may include a display screen (Display), a keyboard (Keyboard), and optionally the user interface 603 may further include a standard wired interface and a wireless interface. The network interface 604 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 605 may be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The memory 605 may optionally also be at least one storage device located far from the aforementioned processor 601. As Figure 6As shown, in the memory 605 serving as a computer-readable storage medium, an operating system, a network communication module, a user interface module, and a device control application program may be included.

[0122] The network interface 604 can provide network communication network elements; the user interface 603 is mainly used to provide an input interface for users; and the processor 601 can be used to call the device control application program stored in the memory 605 to perform the following operations:

[0123] Obtain first sample data corresponding to A data categories respectively, perform feature extraction on the first sample data corresponding to A data categories respectively to obtain A initial feature data; A is a positive integer;

[0124] Perform singular value decomposition on A initial feature data respectively to obtain A first orthogonal matrices, and based on the A first orthogonal matrices, determine principal component matrices corresponding to A data categories respectively;

[0125] Obtain first feature data of second sample data, use the principal component matrices corresponding to A data categories respectively to project the first feature data of the second sample data, obtain category space features corresponding to the second sample data in A data categories respectively, and fuse the A category space features to obtain sample category features;

[0126] Input the sample category features into an initial classification model for prediction to obtain a sample prediction result corresponding to the second sample data, and adjust the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain a target classification model; the target classification model is used for classification prediction among A data categories.

[0127] In a possible implementation manner, the processor 601 is further used to perform the following operations:

[0128] Obtain sample data to be classified, send the sample data to be classified to M service devices for classification processing, and receive initial category labels of the sample data to be classified returned by each service device;

[0129] Integrate the initial category labels of the sample data to be classified respectively by M service devices to obtain a classification category label of the sample data to be classified;

[0130] Based on the classification category label of the sample data to be classified, split the sample data to be classified into first sample data corresponding to A data categories respectively.

[0131] In a possible implementation, the processor 601 performs singular value decomposition on A initial feature data respectively to obtain A first orthogonal matrices, and based on the A first orthogonal matrices, determines the principal component matrices corresponding to A data categories respectively for performing the following operations:

[0132] Perform central offset processing on the initial feature data corresponding to the t-th data category to obtain a first sample matrix, and perform transpose processing on the first sample matrix to obtain a first transposed matrix; t is a positive integer less than or equal to A;

[0133] Perform feature fusion on the first sample matrix and the first transposed matrix to obtain the first orthogonal matrix and the second orthogonal matrix corresponding to the t-th data category;

[0134] Obtain the eigenvalue data of the first orthogonal matrix, and decompose the eigenvalue data to obtain the singular value matrix corresponding to the t-th data category;

[0135] According to the singular value matrix, obtain the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category. When t is equal to A, obtain the principal component matrices corresponding to A data categories respectively.

[0136] In a possible implementation, the processor 601 obtains the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category according to the singular value matrix for performing the following operations:

[0137] Arrange the singular values in the singular value matrix in descending order to obtain a first sequence;

[0138] Obtain a limiting coefficient, and based on the singular values in the singular value matrix and the limiting coefficient, determine a first singular value, and determine a first limiting parameter based on the first sequence and the first singular value; the first limiting parameter is used to make the first singular value located between a first constraint data and a second constraint data; the first constraint data is obtained by processing the first sequence with the first limiting parameter, the second constraint data is obtained by processing the first sequence with a second limiting parameter, and the second limiting parameter is updated from the first limiting parameter;

[0139] Based on the first limiting parameter, determine L columns of data in the first orthogonal matrix corresponding to the t-th data category as the principal component matrix corresponding to the t-th data category; L is a positive integer and L is the first limiting parameter.

[0140] In a possible implementation, the processor 601 projects the first feature data of the second sample data respectively by using the principal component matrices corresponding to A data categories, obtains the category space features corresponding to the second sample data in the A data categories respectively, and performs feature splicing and fusion on the first projection vectors of the A category space features to obtain the sample category features for performing the following operations:

[0141] In the t-th data category, the first feature data of the second sample data is subjected to central offset to obtain the offset feature of the second sample data in the t-th data category;

[0142] The principal component matrix corresponding to the t-th data category is used to perform spatial mapping on the offset feature of the second sample data in the t-th data category to obtain the category space feature of the second sample data in the t-th data category;

[0143] When the category space features corresponding to the second sample data in the A data categories are obtained, the category space features corresponding to the second sample data in the A data categories are subjected to feature fusion to obtain the sample category features corresponding to the second sample data.

[0144] In a possible implementation, the sample prediction result includes the category probabilities for the A data categories of the second sample data; the processor 601 adjusts the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain the target classification model for performing the following operations:

[0145] Obtain the data category label corresponding to the second sample data, and generate a loss function according to the data category label and the A category probabilities;

[0146] The loss function is used to adjust the parameters of the initial classification model until the parameters converge to obtain the target classification model.

[0147] In a possible implementation, the processor 601 is further used to perform the following operations:

[0148] Obtain the data to be predicted, extract features from the data to be predicted to obtain the second feature data, project the second feature data by using the principal component matrices corresponding to the A data categories respectively to obtain the to-be-predicted space features corresponding to the data to be predicted in the A data categories respectively, and perform feature fusion on the A to-be-predicted space features to obtain the to-be-predicted category features;

[0149] Input the to-be-predicted category features into the target classification model for prediction to obtain the prediction probabilities corresponding to the data to be predicted in the A data categories respectively, and determine the data category with the largest prediction probability as the classification prediction result of the data to be predicted.

[0150] In addition, it should be noted here that: The embodiments of the present application further provide a computer-readable storage medium, which stores a computer program, and the computer program is adapted to be loaded and executed by the processor Figure 3 or Figure 4 the methods provided in each step in Figure 3 or Figure 4 The implementation manners provided in each step in can be specifically referred to, and will not be elaborated here. In addition, the description of the beneficial effects of adopting the same method will not be elaborated either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application. As an example, the computer program can be deployed to be executed on a computer device, or on multiple computer devices located at one place, or on multiple computer devices distributed at multiple places and interconnected through a communication network

[0151] The computer-readable storage medium can be the device provided in any of the foregoing embodiments or the internal storage unit of the computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store the data that has been output or will be output

[0152] The embodiments of the present application further provide a computer program product or a computer program, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 3 or Figure 4 the methods provided in various alternative manners in, and therefore, will not be elaborated here

[0153] In the description, claims, and accompanying drawings of the embodiments of this application, terms such as "first" and "second" are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units is not limited to the listed steps or modules, but may optionally further include unlisted steps or modules, or may optionally further include other step units inherent to these processes, methods, devices, products, or equipment.

[0154] In the embodiments of this application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, which works with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the function of that module or unit.

[0155] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in this description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0156] The methods and related devices provided in the embodiments of this application are described with reference to the method flowcharts and / or structural schematic diagrams provided in the embodiments of this application. Specifically, each process and / or block of the method flowchart and / or structural schematic diagram, as well as the combination of the 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 the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable devices generate for implementation in the process Figure 1 a process or multiple processes and / or structural schematic Figure 1a device with the functions specified in one or more boxes. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including the instruction device, or are transmitted via a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wirelessly (e.g., infrared, wireless, microwave, etc.). The instruction device implements the steps in the process Figure 1 one process or multiple processes and / or the structural schematic Figure 1 a device with the functions specified in one or more boxes. These computer program instructions can also be loaded onto a computer or other programmable device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more boxes in one process or multiple processes and / or the structural schematic Figure 1 one process or multiple processes and / or the steps of the functions specified in one or more boxes in the structural schematic.

[0157] The steps in the method of the embodiments of the present application can be adjusted, combined, and deleted according to actual needs.

[0158] The modules in the device of the embodiments of the present application can be combined, divided, and deleted according to actual needs.

[0159] The foregoing disclosure is only for the preferred embodiments of the present application, and of course cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A data processing method, characterized in that, The method includes: Obtaining first sample data corresponding to A data categories respectively, performing feature extraction on the first sample data corresponding to the A data categories respectively to obtain A initial feature data; A is a positive integer; Performing singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and determining principal component matrices corresponding to the A data categories respectively based on the A first orthogonal matrices; Obtaining first feature data of second sample data, projecting the first feature data of the second sample data by using the principal component matrices corresponding to the A data categories respectively to obtain category space features corresponding to the second sample data in the A data categories respectively, and performing feature fusion on the A category space features to obtain sample category features; Inputting the sample category features into an initial classification model for prediction to obtain a sample prediction result corresponding to the second sample data, and adjusting parameters of the initial classification model by using the sample prediction result and a data category label of the second sample data to obtain a target classification model; the target classification model is used for classification prediction among the A data categories.

2. The method according to claim 1, characterized in that, The step of performing singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and determining principal component matrices corresponding to the A data categories respectively based on the A first orthogonal matrices includes: Performing center offset processing on the initial feature data corresponding to the t-th data category to obtain a first sample matrix, and performing transpose processing on the first sample matrix to obtain a first transposed matrix; t is a positive integer less than or equal to A; Performing feature fusion on the first sample matrix and the first transposed matrix to obtain a first orthogonal matrix and a second orthogonal matrix corresponding to the t-th data category; Obtaining eigenvalue data of the first orthogonal matrix, and decomposing the eigenvalue data to obtain a singular value matrix corresponding to the t-th data category; According to the singular value matrix, obtaining the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category, and when t is equal to A, obtaining principal component matrices corresponding to the A data categories respectively.

3. The method according to claim 2, characterized in that The step of obtaining the principal component matrix corresponding to the t-th data category from the first orthogonal matrix corresponding to the t-th data category according to the singular value matrix includes: Arranging the singular values in the singular value matrix in descending order to obtain a first sequence; Obtaining a constraint coefficient, determining a first singular value based on the singular values in the singular value matrix and the constraint coefficient, and determining a first constraint parameter based on the first sequence and the first singular value; the first constraint parameter is used to make the first singular value located between a first constraint data and a second constraint data; the first constraint data is obtained by processing the first sequence by the first constraint parameter, the second constraint data is obtained by processing the first sequence by a second constraint parameter, and the second constraint parameter is updated from the first constraint parameter; Based on the first constraint parameter, determine L columns of data in the first orthogonal matrix corresponding to the t-th data category as the principal component matrix corresponding to the t-th data category; L is a positive integer and L is the first constraint parameter.

4. The method according to claim 1, characterized in that The method of using the principal component matrices corresponding to the A data categories respectively to project the first feature data of the second sample data to obtain the category space features corresponding to the second sample data in the A data categories respectively, and performing feature splicing and fusion on the first projection vectors of the A category space features to obtain the sample category features includes: In the t-th data category, perform central offset on the first feature data of the second sample data to obtain the offset feature of the second sample data in the t-th data category; Use the principal component matrix corresponding to the t-th data category to perform spatial mapping on the offset feature of the second sample data in the t-th data category to obtain the category space feature of the second sample data in the t-th data category; When the category space features corresponding to the second sample data in the A data categories are obtained respectively, perform feature fusion on the category space features corresponding to the second sample data in the A data categories respectively to obtain the sample category feature corresponding to the second sample data.

5. The method according to claim 1, characterized in that, The sample prediction result includes the category probabilities of the A data categories for the second sample data; the method of adjusting the parameters of the initial classification model through the sample prediction result and the data category label of the second sample data to obtain the target classification model includes: Obtain the data category label corresponding to the second sample data, and generate a loss function according to the data category label and the A category probabilities; Use the loss function to adjust the parameters of the initial classification model until the parameters converge to obtain the target classification model.

6. The method according to claim 1, characterized in that, It further includes: Obtain the data to be predicted, extract features from the data to be predicted to obtain second feature data, use the principal component matrices corresponding to the A data categories respectively to project the second feature data to obtain the to-be-predicted space features corresponding to the data to be predicted in the A data categories respectively, and perform feature fusion on the A to-be-predicted space features to obtain the to-be-predicted category feature; Input the to-be-predicted category feature into the target classification model for prediction to obtain the prediction probabilities corresponding to the data to be predicted in the A data categories respectively, and determine the data category with the largest prediction probability as the classification prediction result of the data to be predicted.

7. The method according to claim 1, characterized in that, The method of obtaining the sample data corresponding to the A data categories respectively includes: Obtain the sample data to be classified, send the sample data to be classified to M service devices for classification processing, and receive the initial category labels of the sample data to be classified returned by each service device; Integrate the initial category labels of the sample data to be classified by the M service devices respectively to obtain the classification category label of the sample data to be classified. Based on the classification category labels of the sample data to be classified, the sample data to be classified is split into first sample data corresponding to A data categories respectively.

8. A data processing device, characterized in that, The device includes: A data acquisition module, configured to acquire first sample data corresponding to A data categories respectively; A feature extraction module, configured to perform feature extraction on the first sample data corresponding to the A data categories respectively to obtain A initial feature data; A is a positive integer; A data decomposition module, configured to perform singular value decomposition on the A initial feature data respectively to obtain A first orthogonal matrices, and determine principal component matrices corresponding to the A data categories respectively based on the A first orthogonal matrices; A data processing module, configured to acquire first feature data of second sample data, project the first feature data of the second sample data by using the principal component matrices corresponding to the A data categories respectively to obtain category space features corresponding to the second sample data in the A data categories respectively, and perform feature fusion on the A category space features to obtain sample category features; A model training module, configured to input the sample category features into an initial classification model for prediction to obtain a sample prediction result corresponding to the second sample data, and adjust parameters of the initial classification model based on the sample prediction result and the data category label of the second sample data to obtain a target classification model; the target classification model is used for classification prediction among the A data categories.

9. A computer device, characterized in that, It includes a processor, a memory, and an input / output interface; The processor is respectively connected to the memory and the input / output interface, wherein the input / output interface is used for receiving and outputting data, the memory is used for storing a computer program, and the processor is used for calling the computer program so that the computer device executes the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the method according to any one of claims 1-7.

11. A computer program product, comprising a computer program / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the method according to any one of claims 1-7.

Images