Data intelligent governance system and method based on machine learning

Abstract

The present application relates to the field of intelligent data analysis, and provides a data intelligent governance system and method based on machine learning, which obtains a collection of historical governance data marked as normal data and data to be evaluated, and uses a data processing and analysis algorithm based on machine learning to semantically embed the collection of historical governance data marked as normal data and the data to be evaluated, so as to automatically determine whether the data to be evaluated is normal data based on the semantic interaction matching features between the collection of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in a data set, reduce the risk of misreporting normal data as abnormal, and reduce the underreporting of real abnormal data, thereby improving the quality of the data.

Description

Technical Field

[0001] The present application relates to the field of intelligent data analysis, and more specifically, to a data intelligent governance system and method based on machine learning. Background Art

[0002] High-quality data is the basis for making informed business decisions. Accurate and timely data can help organizations better understand market and business trends. In particular, in many industries, data accuracy and completeness are key to complying with regulatory requirements. Inadequate data quality may lead to legal and compliance issues.

[0003] However, in the current big data governance decision-making process, there are some problems, such as imperfect data sources, incomplete data cleaning, unreasonable parameter settings, and data drift. These problems may cause normal data to be misreported as abnormal or real data problems to be missed, resulting in low data quality, which in turn affects the final decision.

[0004] Therefore, a data intelligence governance solution based on machine learning is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a data intelligent governance system and method based on machine learning, which obtains a set of historical governance data marked as normal data and the data to be evaluated, and uses a data processing and analysis algorithm based on machine learning to semantically embed the set of historical governance data marked as normal data and the data to be evaluated, so as to automatically judge whether the data to be evaluated is normal data based on the semantic interaction matching features between the set of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in the data set, reduce the risk of falsely reporting normal data as abnormal, and reduce the omission of real abnormal data, thereby improving the quality of the data.

[0006] According to one aspect of the present application, a data intelligent governance system based on machine learning is provided, which includes:

[0007] The historical governance data collection module is used to obtain a collection of historical governance data marked as normal data;

[0008] A historical governance data semantic embedding coding module, used for performing semantic embedding coding on each historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors;

[0009] A normal data semantics enhancement module, used for inputting the set of normal data semantics embedding coding vectors into a contextual semantics enhancement module based on saliency-globality to obtain a set of enhanced normal data semantics embedding coding vectors;

[0010] A module for acquiring data to be evaluated, used for acquiring data to be evaluated;

[0011] A semantic embedding coding module for data to be evaluated, used for performing semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector for the data to be evaluated;

[0012] A data one-way interaction module, used for taking the semantic embedding coding vector of the data to be evaluated as a query feature vector, and inputting the query feature vector and the set of the enhanced normal data semantic embedding coding vector into a one-way interaction matching module to obtain a data one-way matching feature vector as a data one-way matching feature;

[0013] The governance result generation module is used to obtain the governance result based on the one-way matching characteristics of the data, and the governance result is used to indicate whether the data to be evaluated is normal data.

[0014] In the above-mentioned data intelligent governance system based on machine learning, the normal data semantic reinforcement module includes: a prominent feature calculation unit, which is used to extract the maximum value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a prominent normal data semantic embedding coding vector; a global feature calculation unit, which is used to extract the average value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a global normal data semantic embedding coding vector; a convolution activation unit, which is used to perform convolution coding and feature activation processing on the prominent normal data semantic embedding coding vector and the global normal data semantic embedding coding vector to obtain a prominent feature normal data semantic embedding coding activation vector and a global feature normal data semantic embedding coding vector. A salient global fusion unit is used to fuse the salient feature normal data semantic embedding coding activation vector and the global feature normal data semantic embedding coding activation vector to obtain a global-salient feature normal data semantic embedding coding vector; a non-linear activation unit is used to perform non-linear activation on the global-salient feature normal data semantic embedding coding vector to obtain a global-salient feature normal data semantic embedding coding weight feature vector; a feature weighted fusion unit is used to use the global-salient feature normal data semantic embedding coding weight feature vector as a weight, perform positional point multiplication on the set of normal data semantic embedding coding vectors and add the set of normal data semantic embedding coding vectors to obtain the set of enhanced normal data semantic embedding coding vectors.

[0015] In the above-mentioned data intelligent governance system based on machine learning, the convolution activation unit includes: a prominent normal data semantic activation feature extraction subunit, which is used to input the prominent normal data semantic embedding convolution coding vector obtained after one-dimensional convolution coding of the prominent normal data semantic embedding coding vector into the ReLU function to obtain a prominent normal data semantic embedding activation coding feature vector; a prominent normal data semantic activation feature enhancement subunit, which is used to perform matrix multiplication of the prominent point convolution feature vector obtained after point convolution coding of the prominent normal data semantic embedding activation coding feature vector with the first weight matrix to obtain the prominent feature normal data semantic embedding coding activation vector; a global normal data semantic activation feature extraction subunit, which is used to input the global normal data semantic embedding convolution coding vector obtained after one-dimensional convolution coding of the global normal data semantic embedding coding vector into the ReLU function to obtain the global normal data semantic embedding activation coding feature vector; a global normal data semantic activation feature enhancement subunit, which is used to perform matrix multiplication of the global point convolution feature vector obtained after point convolution coding of the global normal data semantic embedding activation coding feature vector with the second weight matrix to obtain the global feature normal data semantic embedding coding activation vector.

[0016] In the above-mentioned data intelligent governance system based on machine learning, the nonlinear activation unit is used to: input the global-prominent feature normal data semantic embedding coding vector into the tanh function to obtain a first activated global-prominent feature normal data semantic embedding coding vector; input the global-prominent feature normal data semantic embedding coding vector into the sigmoid function to obtain a second activated global-prominent feature normal data semantic embedding coding vector; perform position point multiplication on the first activated global-prominent feature normal data semantic embedding coding vector and the second activated global-prominent feature normal data semantic embedding coding vector to obtain the global-prominent feature normal data semantic embedding coding weight feature vector.

[0017] In the above-mentioned data intelligent governance system based on machine learning, the data single-item interaction module includes: a semantic association calculation unit, which is used to calculate the semantic association between the query feature vector and each enhanced normal data semantic embedding coding vector in the set of enhanced normal data semantic embedding coding vectors to obtain a set of semantic associations between the data to be evaluated and normal data; and a one-way interactive matching unit, which is used to calculate the vector weighted sum of the set of enhanced normal data semantic embedding coding vectors using the set of semantic associations between the data to be evaluated and normal data as a weight set to obtain the data one-way matching feature vector.

[0018] In the above-mentioned data intelligent governance system based on machine learning, the semantic association calculation unit is used to: calculate the position-by-position difference between the query feature vector and each of the enhanced normal data semantic embedding coding vectors to obtain a set of data semantic difference coding vectors; calculate the norm of each data semantic difference coding vector in the set of data semantic difference coding vectors to obtain a set of semantic associations of the data to be evaluated - normal data.

[0019] In the above-mentioned data intelligent governance system based on machine learning, the governance result generation module is used to: input the data one-way matching feature vector into the classifier-based governance result generation module to obtain the governance result.

[0020] In the above-mentioned machine learning-based data intelligent governance system, it also includes a training module for training the significance-global contextual semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module.

[0021] In the above-mentioned data intelligent governance system based on machine learning, the training module includes: a training data acquisition unit, which is used to acquire training data, wherein the training data includes a set of training history governance data marked as normal data, training data to be evaluated, and a real governance result, wherein the real governance result is a real value of whether the training data to be evaluated is normal data; a training history governance data semantic embedding coding unit, which is used to perform semantic embedding coding on each training history governance data marked as normal data in the set of training history governance data marked as normal data to obtain a set of training normal data semantic embedding coding vectors; a training normal data context semantic reinforcement unit, which is used to input the set of training normal data semantic embedding coding vectors into the context semantic reinforcement module based on significance-globality to obtain a set of training reinforced normal data semantic embedding coding vectors; a training data to be evaluated semantic embedding coding unit, which is used to perform semantic embedding coding on the training data to be evaluated. The evaluation data is semantically embedded and encoded to obtain a semantic embedded encoding vector of the training data to be evaluated; a one-way matching feature generation unit for training data is used to use the semantic embedded encoding vector of the training data to be evaluated as a query feature vector, and input the query feature vector and the set of the training enhanced normal data semantic embedded encoding vector into the one-way interactive matching module to obtain a one-way matching feature vector of the training data; a training governance result generation unit is used to input the one-way matching feature vector of the training data into the classifier-based governance result generation module to obtain a training governance result; a classification loss unit is used to calculate the cross entropy loss function value between the training governance result and the true governance result to obtain a classification loss function value; a model training unit is used to train the saliency-global contextual semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module based on the classification loss function value and through back propagation of gradient descent.

[0022] According to another aspect of the present application, a data intelligent governance method based on machine learning is provided, which includes:

[0023] Obtain a collection of historical governance data that is marked as normal data;

[0024] Performing semantic embedding coding on each piece of historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors;

[0025] Inputting the set of normal data semantic embedding coding vectors into a contextual semantics enhancement module based on saliency-globality to obtain a set of enhanced normal data semantic embedding coding vectors;

[0026] Obtain the data to be evaluated;

[0027] Performing semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector of the data to be evaluated;

[0028] Taking the semantic embedding coding vector of the data to be evaluated as a query feature vector, inputting the query feature vector and the set of the enhanced normal data semantic embedding coding vector into a one-way interactive matching module to obtain a data one-way matching feature vector as a data one-way matching feature;

[0029] Based on the one-way matching characteristics of the data, a governance result is obtained, and the governance result is used to indicate whether the data to be evaluated is normal data.

[0030] Compared with the prior art, the machine learning-based data intelligent governance system and method provided by the present application obtains a collection of historical governance data and data to be evaluated that are marked as normal data, and uses a data processing and analysis algorithm based on machine learning to semantically embed the collection of historical governance data and the data to be evaluated, thereby automatically judging whether the data to be evaluated is normal data based on the semantic interaction matching features between the collection of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in a data set, reduce the risk of misreporting normal data as abnormal, and reduce the omission of real abnormal data, thereby improving the quality of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0032] Figure 1 This is a system block diagram of a data intelligent governance system based on machine learning according to an embodiment of the present application.

[0033] Figure 2 Schematic diagram of the architecture of a data intelligent governance system based on machine learning according to an embodiment of the present application.

[0034] Figure 3 This is a block diagram of a normal data semantic enhancement module in a data intelligent governance system based on machine learning according to an embodiment of the present application.

[0035] Figure 4 It is a block diagram of a convolutional activation unit in a data intelligent management system based on machine learning according to an embodiment of the present application.

[0036] Figure 5It is a block diagram of the data single-item interaction module in the data intelligent governance system based on machine learning according to an embodiment of the present application.

[0037] Figure 6 It is a block diagram of a training module in a data intelligent management system based on machine learning according to an embodiment of the present application.

[0038] Figure 7 This is a flowchart of a data intelligent governance method based on machine learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0040] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0041] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0042] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0043] High-quality data assets are the cornerstone of making smart business decisions. Their accuracy and timeliness are crucial for gaining insight into market and business trends. In many fields, ensuring the accuracy and completeness of data is the only way to meet regulatory requirements. Data quality defects may lead to legal and compliance challenges.

[0044] There are several challenges in the current implementation of big data governance and decision-making, including but not limited to defects in data sources, insufficient data cleaning, unreasonable parameter configuration, and data drift. Due to the potential risks of these problems, data anomalies that should have been correctly identified are ignored, or real data problems are missed, which ultimately affects the reliability of the data and may have an adverse impact on decision-making. Therefore, an optimized data intelligent governance solution is expected.

[0045] As you can understand, machine learning is a branch of artificial intelligence that enables computers to learn through experience and improve their ability to perform tasks without being explicitly programmed. This field combines statistics, algorithm theory, and in-depth analysis of data to learn and extract useful patterns and knowledge from large amounts of data.

[0046] Based on this, the present application proposes a data intelligent governance solution based on machine learning, which obtains a collection of historical governance data marked as normal data and the data to be evaluated, and uses a data processing and analysis algorithm based on machine learning to semantically embed the collection of historical governance data marked as normal data and the data to be evaluated, so as to automatically determine whether the data to be evaluated is normal data based on the semantic interaction matching features between the collection of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in the data set, reduce the risk of misreporting normal data as abnormal, and reduce the omission of real abnormal data, thereby improving the quality of the data.

[0047] Figure 1 This is a system block diagram of a data intelligent governance system based on machine learning according to an embodiment of the present application. Figure 2 Schematic diagram of the architecture of the data intelligent management system based on machine learning according to the embodiment of the present application. Figure 1 and Figure 2 As shown, in the data intelligent governance system 100 based on machine learning, it includes: a historical governance data acquisition module 110, which is used to obtain a set of historical governance data marked as normal data; a historical governance data semantic embedding coding module 120, which is used to semantically embed the historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors; a normal data semantic reinforcement module 130, which is used to input the set of normal data semantic embedding coding vectors into a contextual semantic reinforcement module based on significance-globality to obtain a set of reinforced normal data semantic embedding coding vectors; to be evaluated The data acquisition module 140 is used to acquire the data to be evaluated; the semantic embedding coding module 150 of the data to be evaluated is used to perform semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector of the data to be evaluated; the data unidirectional interaction module 160 is used to use the semantic embedding coding vector of the data to be evaluated as a query feature vector, and input the query feature vector and the set of the enhanced normal data semantic embedding coding vector into the unidirectional interaction matching module to obtain a data unidirectional matching feature vector as a data unidirectional matching feature; the governance result generation module 170 is used to obtain a governance result based on the data unidirectional matching feature, and the governance result is used to indicate whether the data to be evaluated is normal data.

[0048] In an embodiment of the present application, the historical governance data acquisition module 110 is used to obtain a set of historical governance data marked as normal data. It should be understood that the set of historical governance data marked as normal data provides a baseline for normal data behavior. By analyzing these data, the machine learning model can learn the characteristics and patterns of normal data, so that the data to be evaluated in the data set can be more accurately identified to determine whether it is normal data. In particular, in a specific embodiment of the present application, the set of historical governance data marked as normal data can be obtained by marking and sorting the historical governance data.

[0049] In an embodiment of the present application, the historical governance data semantic embedding coding module 120 is used to perform semantic embedding coding on each historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors. Accordingly, considering that each historical governance data marked as normal data in the set of historical governance data marked as normal data contains historical governance data of different times or types, these data contain semantic information and associations. Therefore, in the technical solution of the present application, semantic embedding coding is performed on each historical governance data marked as normal data in the set of historical governance data marked as normal data to convert it into a set of dense normal data semantic embedding coding vectors, thereby better characterizing the semantic features of each historical governance data. In particular, in a specific embodiment of the present application, each historical governance data marked as normal data can be semantically embedded and coded by using a learnable embedding matrix of an embedding layer to obtain a set of normal data semantic embedding coding vectors.

[0050] In an embodiment of the present application, the normal data semantic enhancement module 130 is used to input the set of normal data semantic embedding coding vectors into a contextual semantic enhancement module based on significance-globality to obtain a set of enhanced normal data semantic embedding coding vectors. Accordingly, it is considered that the set of normal data semantic embedding coding vectors reflects significant key normal data semantic association information and global semantic information based on the entire normal data set. Therefore, in order to further enhance the normal data semantic embedding coding vectors by utilizing the contextual semantic relationship between normal data and the global semantic information, in the technical solution of the present application, the set of normal data semantic embedding coding vectors is input into a contextual semantic enhancement module based on significance-globality to obtain a set of enhanced normal data semantic embedding coding vectors.

[0051] In detail, the contextual semantic enhancement module based on saliency-globality calculates the prominent features and global features in the set of normal data semantic embedding coding vectors respectively to capture and mine the most important significant semantic association information in the entire data set and the global semantic association and characteristics of the entire data set. Then, the obtained prominent features and global features are convolutionally encoded and nonlinearly processed respectively to improve the expressiveness and semantic richness of the normal data semantic features. Then, the features obtained after nonlinear processing of the two are fused so that the model can more comprehensively consider the semantic information of different aspects in the set of normal data. Finally, the normal data prominent global features obtained after fusion are nonlinearly activated to more deeply capture the complex semantic information implied in the fusion features, and these complex features are used as weights to weight the set of normal data semantic embedding coding vectors vector by vector and fuse the set of normal data semantic embedding coding vectors to semantically optimize and enhance each of the normal data semantic embedding coding vectors, so that these vectors have more distinguishing and representational capabilities, thereby further improving the processing and analysis efficiency of normal data by the data intelligent governance system.

[0052] Specifically, in the embodiments of the present application, Figure 3 FIG. 1 is a block diagram of a normal data semantic enhancement module in a data intelligent management system based on machine learning according to an embodiment of the present application. Figure 3As shown, the normal data semantic enhancement module 130 includes: a prominent feature calculation unit 131, which is used to extract the maximum value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a prominent normal data semantic embedding coding vector; a global feature calculation unit 132, which is used to extract the average value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a global normal data semantic embedding coding vector; a convolution activation unit 133, which is used to perform convolution coding and feature activation processing on the prominent normal data semantic embedding coding vector and the global normal data semantic embedding coding vector to obtain a prominent feature normal data semantic embedding coding activation vector and a global feature normal data semantic embedding coding activation vector ; A prominent global fusion unit 134 is used to fuse the prominent feature normal data semantic embedding coding activation vector and the global feature normal data semantic embedding coding activation vector to obtain a global-prominent feature normal data semantic embedding coding vector; a non-linear activation unit 135 is used to perform non-linear activation on the global-prominent feature normal data semantic embedding coding vector to obtain a global-prominent feature normal data semantic embedding coding weight feature vector; a feature weighted fusion unit 136 is used to use the global-prominent feature normal data semantic embedding coding weight feature vector as a weight, perform positional dot multiplication on the set of normal data semantic embedding coding vectors and add the set of normal data semantic embedding coding vectors to obtain the set of enhanced normal data semantic embedding coding vectors.

[0053] More specifically, in the embodiments of the present application, Figure 4 FIG. 1 is a block diagram of a convolution activation unit in a data intelligent management system based on machine learning according to an embodiment of the present application. Figure 4As shown, the convolution activation unit 133 includes: a prominent normal data semantic activation feature extraction subunit 1331, which is used to input the prominent normal data semantic embedding convolution coding vector obtained after performing one-dimensional convolution coding on the prominent normal data semantic embedding coding vector into the ReLU function to obtain a prominent normal data semantic embedding activation coding feature vector; a prominent normal data semantic activation feature enhancement subunit 1332, which is used to perform matrix multiplication of the prominent point convolution feature vector obtained after performing point convolution coding on the prominent normal data semantic embedding activation coding feature vector and the first weight matrix to obtain the prominent feature normal data semantic embedding activation coding feature vector. Embedded coding activation vector; a global normal data semantic activation feature extraction subunit 1333, used to input the global normal data semantic embedding convolution coding vector obtained after one-dimensional convolution coding of the global normal data semantic embedding coding vector into the ReLU function to obtain a global normal data semantic embedding activation coding feature vector; a global normal data semantic activation feature enhancement subunit 1334, used to perform matrix multiplication of the global point convolution feature vector obtained after point convolution coding of the global normal data semantic embedding activation coding feature vector with a second weight matrix to obtain the global feature normal data semantic embedding coding activation vector.

[0054] More specifically, in an embodiment of the present application, the nonlinear activation unit is used to: input the global-salient feature normal data semantic embedding coding vector into a tanh function to obtain a first activated global-salient feature normal data semantic embedding coding vector; input the global-salient feature normal data semantic embedding coding vector into a sigmoid function to obtain a second activated global-salient feature normal data semantic embedding coding vector; and perform position point multiplication on the first activated global-salient feature normal data semantic embedding coding vector and the second activated global-salient feature normal data semantic embedding coding vector to obtain the global-salient feature normal data semantic embedding coding weight feature vector.

[0055] In an embodiment of the present application, specifically, the normal data semantic enhancement module is used to: use the contextual semantic enhancement module based on saliency-globality to process the set of normal data semantic embedding coding vectors with the following semantic enhancement formula to obtain the set of enhanced normal data semantic embedding coding vectors; wherein the semantic enhancement formula is:

[0056]

[0057]

[0058] C=tanh(C1+C2)⊙σ(C1+C2)

[0059] X'=X⊙C+X

[0060] Where X is the set of normal data semantic embedding coding vectors, Max(·) and Avg(·) are the maximum value and average value of each feature vector in the set of feature vectors, Conv(·) is a one-dimensional convolutional coding, ReLU is the ReLU function, Conv 1×1 (·) is the point convolutional coding, and are the first weight matrix and the second weight matrix respectively, C1 and C2 are the salient feature normal data semantic embedding coding activation vector and the global feature normal data semantic embedding coding activation vector respectively, σ(·) is the Sigmoid function, tanh is the tanh function, ⊙ represents the positional point multiplication, C is the global-salient feature normal data semantic embedding coding weight feature vector, and X' is the set of enhanced normal data semantic embedding coding vectors.

[0061] In the embodiment of the present application, the data acquisition module 140 is used to acquire the data to be evaluated. It should be understood that the data to be evaluated is data that needs to be tested and verified. By processing and analyzing the data to be evaluated, it can be determined whether it is abnormal, so that the model can accurately find the abnormal data in the data to be evaluated and realize automatic identification of the abnormal data.

[0062] In an embodiment of the present application, the semantic embedding coding module 150 of the data to be evaluated is used to perform semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector of the data to be evaluated. It should be understood that the data to be evaluated also contains a large amount of semantic information about the data to be evaluated, and semantic embedding coding can reduce the dimension of the data while capturing the semantic information in the data. Therefore, in the technical solution of the present application, semantic embedding coding is performed on the data to be evaluated to convert the data into a semantic embedding coding vector of the data to be evaluated containing its rich semantics. In particular, in a specific embodiment of the present application, a learnable embedding matrix of an embedding layer can be used to perform semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector of the data to be evaluated.

[0063] In an embodiment of the present application, the data single item interaction module 160 is used to use the semantic embedding coding vector of the data to be evaluated as a query feature vector, and input the query feature vector and the set of the enhanced normal data semantic embedding coding vector into the one-way interaction matching module to obtain the data one-way matching feature vector as the data one-way matching feature. Accordingly, in order to match the semantic embedding coding vector of the data to be evaluated with the set of the enhanced normal data semantic embedding coding vector, so as to calculate the similarity or matching measure between the data to be evaluated and the normal data, a more accurate judgment can be made on the data to be evaluated. Therefore, in the technical solution of the present application, the semantic embedding coding vector of the data to be evaluated is used as a query feature vector, and the query feature vector and the set of the enhanced normal data semantic embedding coding vector are input into the one-way interaction matching module to obtain the data one-way matching feature vector. Specifically, the one-way interaction matching module first calculates the semantic similarity between the query feature vector and each enhanced normal data semantic embedding coding vector to measure the degree of semantic association matching between the data to be evaluated and each enhanced normal data semantic embedding coding vector. Then, each semantic similarity is used as a weight value to weight and fuse each of the enhanced normal data semantic embedding coding vectors to obtain a data one-way matching feature vector. In this way, the semantic similarity interaction matching degree between the data to be evaluated and the normal data can be more accurately reflected, and the accuracy of subsequent abnormal judgment of the data to be evaluated can be improved.

[0064] Specifically, in the embodiments of the present application, Figure 5 FIG. 1 is a block diagram of a data single-item interaction module in a data intelligent management system based on machine learning according to an embodiment of the present application. Figure 5 As shown, the data single-item interaction module 160 includes: a semantic association calculation unit 161, which is used to calculate the semantic association between the query feature vector and each enhanced normal data semantic embedding coding vector in the set of enhanced normal data semantic embedding coding vectors to obtain a set of semantic associations of the data to be evaluated-normal data; and a unidirectional interaction matching unit 162, which is used to calculate the vector weighted sum of the set of enhanced normal data semantic embedding coding vectors using the set of semantic associations of the data to be evaluated-normal data as a weight set to obtain the data unidirectional matching feature vector.

[0065] More specifically, in an embodiment of the present application, the semantic association calculation unit is used to: calculate the position-by-position difference between the query feature vector and each of the enhanced normal data semantic embedding coding vectors to obtain a set of data semantic difference coding vectors; calculate the norm of each data semantic difference coding vector in the set of data semantic difference coding vectors to obtain a set of semantic associations of the data to be evaluated - normal data.

[0066] In an embodiment of the present application, specifically, the data unidirectional interaction module is used to: use the semantic embedding coding vector of the data to be evaluated as the query feature vector, and use the unidirectional interaction matching module to process the query feature vector and the set of the enhanced normal data semantic embedding coding vector using the following unidirectional matching formula to obtain a data unidirectional matching feature vector; wherein the unidirectional matching formula is:

[0067] r i =‖v a -v i ‖1

[0068]

[0069] Among them, v a is the query feature vector, v i is the i-th enhanced normal data semantic embedding coding vector in the set of enhanced normal data semantic embedding coding vectors, ‖·‖1 represents the one-norm of the feature vector, r i represents the i-th semantic association degree between the data to be evaluated and the normal data in the set of semantic association degrees between the data to be evaluated and the normal data, ⊙ represents the point multiplication by position, n is the number of feature vectors in the set of enhanced normal data semantic embedding coding vectors, and s is the one-way matching feature vector of the data.

[0070] In an embodiment of the present application, the governance result generation module 170 is used to obtain a governance result based on the one-way matching feature of the data, and the governance result is used to indicate whether the data to be evaluated is normal data. Specifically, the governance result generation module is used to: input the one-way matching feature vector of the data into the classifier-based governance result generation module to obtain the governance result. That is, the one-way matching feature of the data obtained by one-way interactive matching of the query feature vector and the set of enhanced normal data semantic embedding coding vectors is used for classification processing to automatically determine whether the data to be evaluated is normal data. In this way, it is possible to automatically identify abnormal or non-normal data in the data set, reduce the risk of falsely reporting normal data as abnormal, and at the same time reduce the underreporting of real abnormal data, thereby improving the quality of the data.

[0071] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0072] Specifically, in an embodiment of the present application, it also includes a training module for training the saliency-globality-based contextual semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module.

[0073] Figure 6 FIG. 1 is a block diagram of a training module in a data intelligent management system based on machine learning according to an embodiment of the present application. Figure 6 As shown, the training module 200 includes: a training data acquisition unit 210, which is used to acquire training data, wherein the training data includes a set of training history governance data marked as normal data, training data to be evaluated, and a true governance result, wherein the true governance result is a true value of whether the training data to be evaluated is normal data; a training history governance data semantic embedding coding unit 220, which is used to perform semantic embedding coding on each training history governance data marked as normal data in the set of training history governance data marked as normal data to obtain a set of training normal data semantic embedding coding vectors; a training normal data context semantic reinforcement unit 230, which is used to input the set of training normal data semantic embedding coding vectors into the saliency-globality-based context semantic reinforcement module to obtain a set of training reinforced normal data semantic embedding coding vectors; a training data to be evaluated semantic embedding coding unit 240, which is used to perform semantic embedding coding on the training data to be evaluated. The semantic embedding coding is used to obtain the semantic embedding coding vector of the training data to be evaluated; the one-way matching feature generation unit 250 for training data is used to use the semantic embedding coding vector of the training data to be evaluated as the query feature vector, and input the query feature vector and the set of the training enhanced normal data semantic embedding coding vector into the one-way interactive matching module to obtain the one-way matching feature vector of the training data; the training governance result generation unit 260 is used to input the one-way matching feature vector of the training data into the classifier-based governance result generation module to obtain the training governance result; the classification loss unit 270 is used to calculate the cross entropy loss function value between the training governance result and the true governance result to obtain the classification loss function value; the model training unit 280 is used to train the saliency-global context semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module based on the classification loss function value and through back propagation of gradient descent.

[0074] It should be understood that here, for the set of training normal data semantic embedding coding vectors that express the embedded coding semantic features of each historical governance data, after being enhanced through the saliency-global contextual semantics, there will be a deviation in the semantic feature distribution relative to the initial embedded coding semantic features of the training data to be evaluated semantic embedding coding vector in terms of semantic feature representation, which will lead to the one-way interactive matching class imbalance of the data one-way matching feature vector obtained by one-way interactive matching of the training data to be evaluated semantic embedding coding vector and the set of training enhanced normal data semantic embedding coding vectors, and make the interactive matching logic convergence of the probability density distribution of in-class features and out-of-class features after the clustering operation inconsistent.

[0075] Based on this, in this preferred embodiment, when the contextual semantic enhancement module based on significance-globality, the one-way interactive matching module and the classifier-based governance result generation module are trained based on the classification loss function value, wherein, in each iteration of the training, the one-way matching feature vector of the training data is iteratively optimized,

[0076] Wherein, the iterative optimization process includes: performing clustering operations on the one-way matching feature vectors of the training data, for example, performing clustering operations based on the distance between feature values, and determining the number of in-class features and the number of out-of-class features; dividing the total number of eigenvalues ​​of the one-way matching feature vectors of the training data by the number of in-class features and the number of out-of-class features respectively to obtain a class importance value and an out-of-class proportion value, and calculating the inverse of the class importance value to obtain a class constraint value; calculating a power function of each eigenvalue of the one-way matching feature vector of the training data with the class importance value as an exponent, and adding it to an exponential value with a natural function as a base and the class constraint value as an exponent, and then multiplying it by the out-of-class proportion value to obtain a one-way matching modulation vector of the training data; performing point multiplication of the one-way matching feature vector of the training data with the class importance value to obtain a one-way matching ontology vector of the training data; and calculating the weighted sum of the one-way matching modulation vector of the training data and the one-way matching ontology vector of the training data with a weight hyperparameter to obtain an optimized one-way matching feature vector of the training data.

[0077] Specifically, in this preferred embodiment, the training data one-way matching feature vector is optimized to obtain an optimized training data one-way matching feature vector. The process formula is expressed as follows:

[0078]

[0079] Wherein, V is the one-way matching feature vector of the training data, n is the total number of eigenvalues ​​of the one-way matching feature vector of the training data, k is the number of intra-class features of the one-way matching feature vector of the training data, ⊙ is the point multiplication by position, β is the weight hyperparameter, ⊕ is the addition by position, represents exponential operation on the feature values ​​at each position in the feature vector, and V' is the optimized one-way matching feature vector of the training data.

[0080] Therefore, while performing clustering operations on the one-way matching feature vectors of the training data, the clustering importance measure of the feature value of the one-way matching feature vectors of the training data is combined with the class constraint measure of the clustering operation to further perform responsive modulation with the out-of-class scale factor, and based on the clustering scaling represented by the feature ontology of the one-way matching feature vectors of the training data, the simplicity and effectiveness of the overall feature purification of the feature set of the one-way matching feature vectors of the training data by the clustering operation is established, and the inconsistency of the interactive matching logic convergence of the probability density distribution caused by the clustering operation is suppressed, so as to improve the classification iteration effect of the governance result generation module based on the classifier input of the feature set of the one-way matching feature vectors of the training data, that is, to improve the speed of classification training and the accuracy of classification results. In this way, the abnormal or non-normal data in the data set can be automatically identified, the risk of misreporting normal data as abnormal can be reduced, and the real abnormal data can be omitted, thereby improving the quality of the data.

[0081] In summary, the data intelligent governance system 100 based on machine learning according to the embodiment of the present application is explained, which obtains a set of historical governance data and data to be evaluated that are marked as normal data, and uses a data processing and analysis algorithm based on machine learning to semantically embed the set of historical governance data marked as normal data and the data to be evaluated, so as to automatically determine whether the data to be evaluated is normal data based on the semantic interaction matching features between the set of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in a data set, reduce the risk of misreporting normal data as abnormal, and reduce the omission of real abnormal data, thereby improving the quality of the data.

[0082] As described above, the data intelligent management system 100 based on machine learning according to the embodiment of the present application can be implemented in various terminal devices, such as a server for data intelligent management based on machine learning. In one example, the data intelligent management system 100 based on machine learning according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the data intelligent management system 100 based on machine learning can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the data intelligent management system 100 based on machine learning can also be one of the many hardware modules of the terminal device.

[0083] Alternatively, in another example, the machine learning-based data intelligent management system 100 and the terminal device may also be separate devices, and the machine learning-based data intelligent management system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0084] Figure 7 Flowchart of the data intelligent management method based on machine learning according to the embodiment of the present application. Figure 7 As shown, in the data intelligent governance method based on machine learning, it includes: S110, obtaining a set of historical governance data marked as normal data; S120, semantically embedding and coding each historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding and coding vectors; S130, inputting the set of normal data semantic embedding and coding vectors into a contextual semantic enhancement module based on significance-globality to obtain a set of enhanced normal data semantic embedding and coding vectors; S140, obtaining data to be evaluated; S150, semantically embedding and coding the data to be evaluated to obtain a semantic embedding and coding vector of the data to be evaluated; S160, using the semantic embedding and coding vector of the data to be evaluated as a query feature vector, inputting the query feature vector and the set of enhanced normal data semantic embedding and coding vectors into a one-way interactive matching module to obtain a data one-way matching feature vector as a data one-way matching feature; S170, based on the data one-way matching feature, obtaining a governance result, and the governance result is used to indicate whether the data to be evaluated is normal data.

[0085] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned data intelligent governance method based on machine learning have been referred to above. Figures 1 to 6 The description of the machine learning-based data intelligent governance system has been introduced in detail, and therefore, its repeated description will be omitted.

[0086] In summary, the data intelligent governance method based on machine learning based on the embodiment of the present application is explained, which obtains a set of historical governance data marked as normal data and the data to be evaluated, and uses a data processing and analysis algorithm based on machine learning to semantically embed the set of historical governance data marked as normal data and the data to be evaluated, so as to automatically determine whether the data to be evaluated is normal data based on the semantic interaction matching characteristics between the set of historical governance data and the data to be evaluated. In this way, it is possible to automatically identify abnormal or non-normal data in the data set, reduce the risk of falsely reporting normal data as abnormal, and reduce the omission of real abnormal data, thereby improving the quality of the data.

[0087] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of the technical solution of the present application.

Claims

1. A data intelligent management system based on machine learning, characterized in that: include: The historical governance data collection module is used to obtain a collection of historical governance data marked as normal data; A historical governance data semantic embedding coding module, used for performing semantic embedding coding on each historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors; A normal data semantics enhancement module, used for inputting the set of normal data semantics embedding coding vectors into a contextual semantics enhancement module based on saliency-globality to obtain a set of enhanced normal data semantics embedding coding vectors; A module for acquiring data to be evaluated, used for acquiring data to be evaluated; A semantic embedding coding module for data to be evaluated, used for performing semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector for the data to be evaluated; A data one-way interaction module, used for taking the semantic embedding coding vector of the data to be evaluated as a query feature vector, and inputting the query feature vector and the set of the enhanced normal data semantic embedding coding vector into a one-way interaction matching module to obtain a data one-way matching feature vector as a data one-way matching feature; A governance result generation module, used to obtain a governance result based on the one-way matching feature of the data, wherein the governance result is used to indicate whether the data to be evaluated is normal data; Wherein, the normal data semantic enhancement module includes: A salient feature calculation unit, used for extracting the maximum value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a salient normal data semantic embedding coding vector; A global feature calculation unit, used for extracting the average value of each normal data semantic embedding coding vector in the set of normal data semantic embedding coding vectors to obtain a global normal data semantic embedding coding vector; A convolution activation unit, used for performing convolution coding and feature activation processing on the salient normal data semantic embedding coding vector and the global normal data semantic embedding coding vector to obtain a salient feature normal data semantic embedding coding activation vector and a global feature normal data semantic embedding coding activation vector; A salient global fusion unit, used for fusing the salient feature normal data semantic embedding coding activation vector and the global feature normal data semantic embedding coding activation vector to obtain a global-salient feature normal data semantic embedding coding vector; A non-linear activation unit, used for performing non-linear activation on the global-salient feature normal data semantic embedding coding vector to obtain a global-salient feature normal data semantic embedding coding weight feature vector; A feature weighted fusion unit is used to use the global-salient feature normal data semantic embedding coding weight feature vector as a weight, perform positional dot multiplication on the set of normal data semantic embedding coding vectors and add the set of normal data semantic embedding coding vectors to obtain the set of enhanced normal data semantic embedding coding vectors.

2. The data intelligent management system based on machine learning according to claim 1 is characterized in that: The convolution activation unit comprises: A salient normal data semantic activation feature extraction subunit is used to input the salient normal data semantic embedding convolutional coding vector obtained by performing one-dimensional convolution coding on the salient normal data semantic embedding coding vector into a ReLU function to obtain a salient normal data semantic embedding activation coding feature vector; A salient normal data semantic activation feature enhancement subunit, used for performing matrix multiplication of a salient point convolution feature vector obtained by performing point convolution encoding on the salient normal data semantic embedding activation coding feature vector and a first weight matrix to obtain the salient feature normal data semantic embedding coding activation vector; A global normal data semantic activation feature extraction subunit, used for inputting the global normal data semantic embedding convolutional coding vector obtained after performing one-dimensional convolution coding on the global normal data semantic embedding coding vector into a ReLU function to obtain a global normal data semantic embedding activation coding feature vector; The global normal data semantic activation feature enhancement subunit is used to perform matrix multiplication of the global point convolution feature vector obtained after point convolution encoding the global normal data semantic embedding activation coding feature vector with the second weight matrix to obtain the global feature normal data semantic embedding coding activation vector.

3. The data intelligent management system based on machine learning according to claim 2 is characterized in that: The nonlinear activation unit is used to: Inputting the global-salient feature normal data semantic embedding coding vector into a tanh function to obtain a first activated global-salient feature normal data semantic embedding coding vector; Inputting the global-salient feature normal data semantic embedding coding vector into a Sigmoid function to obtain a second activated global-salient feature normal data semantic embedding coding vector; The first activated global-salient feature normal data semantic embedding coding vector and the second activated global-salient feature normal data semantic embedding coding vector are multiplied by position point to obtain the global-salient feature normal data semantic embedding coding weight feature vector.

4. The data intelligent management system based on machine learning according to claim 3 is characterized in that: The data single-item interaction module includes: A semantic relevance calculation unit, used for calculating the semantic relevance between the query feature vector and each enhanced normal data semantic embedding coding vector in the set of enhanced normal data semantic embedding coding vectors to obtain a set of semantic relevance between the data to be evaluated and the normal data; The one-way interactive matching unit is used to calculate the vector weighted sum of the set of enhanced normal data semantic embedding coding vectors using the set of semantic associations between the data to be evaluated and the normal data as a weight set to obtain the data one-way matching feature vector.

5. The data intelligent management system based on machine learning according to claim 4 is characterized in that: The semantic relevance calculation unit is used to: Calculating the position-by-position difference between the query feature vector and each of the enhanced normal data semantic embedding encoding vectors to obtain a set of data semantic difference encoding vectors; A norm of each data semantic difference encoding vector in the set of data semantic difference encoding vectors is calculated to obtain a set of semantic associations of the data to be evaluated and normal data.

6. The data intelligent management system based on machine learning according to claim 5 is characterized in that: The governance result generation module is used to: input the data one-way matching feature vector into the classifier-based governance result generation module to obtain the governance result.

7. The data intelligent management system based on machine learning according to claim 6 is characterized in that: It also includes a training module for training the saliency-globality-based contextual semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module.

8. The data intelligent management system based on machine learning according to claim 7 is characterized in that: The training module comprises: A training data acquisition unit, configured to acquire training data, wherein the training data includes a set of training historical management data marked as normal data, training data to be evaluated, and a true management result, wherein the true management result is a true value of whether the training data to be evaluated is normal data; A training history governance data semantic embedding coding unit, used for performing semantic embedding coding on each training history governance data marked as normal data in the set of training history governance data marked as normal data to obtain a set of training normal data semantic embedding coding vectors; A training normal data context semantics reinforcement unit, used for inputting the set of training normal data semantic embedding coding vectors into the context semantics reinforcement module based on saliency-globality to obtain a set of training reinforced normal data semantic embedding coding vectors; A semantic embedding coding unit for training data to be evaluated, used for performing semantic embedding coding on the training data to be evaluated to obtain a semantic embedding coding vector for the training data to be evaluated; A training data one-way matching feature generating unit, used for taking the semantic embedding coding vector of the training data to be evaluated as a query feature vector, and inputting the query feature vector and the set of the training enhanced normal data semantic embedding coding vector into the one-way interactive matching module to obtain the training data one-way matching feature vector; A training governance result generating unit, used for inputting the one-way matching feature vector of the training data into the classifier-based governance result generating module to obtain a training governance result; A classification loss unit, used to calculate a cross entropy loss function value between the training governance result and the true governance result to obtain a classification loss function value; A model training unit is used to train the saliency-global contextual semantic enhancement module, the one-way interactive matching module and the classifier-based governance result generation module based on the classification loss function value and through back propagation of gradient descent.

9. A method for intelligent data management based on machine learning, using the intelligent data management system based on machine learning according to claim 1, characterized in that: include: Obtain a collection of historical governance data that is marked as normal data; Performing semantic embedding coding on each piece of historical governance data marked as normal data in the set of historical governance data marked as normal data to obtain a set of normal data semantic embedding coding vectors; Inputting the set of normal data semantic embedding coding vectors into a contextual semantics enhancement module based on saliency-globality to obtain a set of enhanced normal data semantic embedding coding vectors; Obtain the data to be evaluated; Performing semantic embedding coding on the data to be evaluated to obtain a semantic embedding coding vector of the data to be evaluated; Taking the semantic embedding coding vector of the data to be evaluated as a query feature vector, inputting the query feature vector and the set of the enhanced normal data semantic embedding coding vector into a one-way interactive matching module to obtain a data one-way matching feature vector as a data one-way matching feature; Based on the one-way matching characteristics of the data, a governance result is obtained, and the governance result is used to indicate whether the data to be evaluated is normal data.

Images