A method, system, device, and medium for classifying and grading data assets.

Abstract

This application provides a data asset classification and grading method, system, device, and medium. The method includes: acquiring a data table to be graded, wherein the data table includes a table name and field names; classifying the data table according to its table name to establish a mapping relationship between the data table and a preset three-level directory, wherein multiple four-level directories with preset security levels are mounted under the three-level directories, and each four-level directory includes multiple asset items; and calling the corresponding asset item in the four-level directory as a target asset item based on the mapping relationship, and determining the security level of the data table to be graded based on the security level of the target asset item with the highest similarity to the field names of the data table. This method can save costs in various information data management tasks, improve work efficiency, and protect and standardize data resources.

Description

Technical Field

[0001] This invention relates to the field of intelligent data processing, and in particular to a method, system, device, and medium for classifying and grading data assets. Background Technology

[0002] With the development of information technology, financial institutions and financial transactions have achieved informatization and digitalization. The information generated is gradually being transformed into important digital assets of institutions in different forms. Financial data has become richer with the explosive growth of application scenarios and institutions. On the other hand, the impact of security threats such as data leakage of financial institutions has gradually shifted from within institutions to between industries, and may even have a certain impact on social production and national security. Financial data that is not managed in a hierarchical manner faces potential security risks in the process of flowing between different businesses and institutions, from technology to management. Summary of the Invention

[0003] In view of the problems existing in the prior art, the present invention proposes a data asset classification and grading method, system, device and medium, which mainly solves the problem that the existing financial data management has security risks and affects the normal operation of related work in the industry.

[0004] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows.

[0005] This application provides a method for classifying and grading data assets, including:

[0006] Obtain the data table to be classified, wherein the data table to be classified includes the table name and field names;

[0007] The data tables to be classified are classified according to their table names to establish a mapping relationship between the data tables to be classified and the preset three-level directories. The three-level directories are mounted with multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items.

[0008] Based on the mapping relationship, the corresponding asset item in the fourth-level directory is called as the target asset item, so as to determine the security level of the data table to be classified according to the security level of the target asset item with the highest similarity to the field name of the data table to be classified.

[0009] In one embodiment of this application, before classifying the data based on the table name of the data table to be classified, the method further includes:

[0010] Obtain data samples;

[0011] Extract the entities and relationships between entities from the data sample to obtain a knowledge graph of the data asset. The knowledge graph consists of triples of entity-relationship-entity correspondence.

[0012] The model is trained based on the data samples and the knowledge graph to obtain a classification model, which is then used to classify the table names of the data tables to be classified.

[0013] In one embodiment of this application, model training based on the data samples and the knowledge graph includes:

[0014] Convert the data samples into word vectors;

[0015] The triples in the knowledge graph are converted into embedded vector representations of entities and relations using a preset vector transformation model, thus obtaining the embedded vectors.

[0016] The word vectors and the embedding vectors are concatenated and then input into a preset initial classification model to establish a mapping relationship between data samples and different asset items, thereby obtaining the classification model.

[0017] In one embodiment of this application, the initial classification model includes: multiple feature extraction units and a fully connected layer. The input of the initial classification model passes through multiple feature extraction units in sequence, and then passes through the fully connected layer for classification prediction to obtain the probability of different classification categories. The feature extraction unit includes multiple convolutional layers of equal length and a pooling layer.

[0018] In one embodiment of this application, before determining the security level of the data table to be classified based on the security level of the asset item with the highest similarity to the field names of the data table to be classified, the following steps are included:

[0019] Convert the field names of the data table to be classified into a first vector;

[0020] Convert the target asset item into a second vector;

[0021] Calculate the similarity between the first vector and the second vector to determine the target asset item with the highest similarity to the field name.

[0022] In one embodiment of this application, before determining the security level of the data table to be classified based on the security level of the asset item with the highest similarity to the field names of the data table to be classified, the method further includes:

[0023] The field names of the data table to be classified are concatenated with the target asset item to obtain a concatenation vector;

[0024] The concatenated vector is input into a pre-trained matching model to determine the target asset item with the highest similarity to the field name.

[0025] In one embodiment of this application, before classifying the data based on the table name of the data table to be classified, the method further includes:

[0026] Obtain the sample data table, and concatenate the table name of the sample data table with the third-level directory to use as the input sample;

[0027] By replacing the phrases in the input samples with synonyms using a preset thesaurus, and then expanding the input samples to obtain an input sample set;

[0028] A pre-classification model is trained based on the input sample set to classify the table names of the data table to be classified based on the pre-classification model.

[0029] This application also provides a data asset classification and grading system, including:

[0030] The data acquisition module is used to acquire the data table to be classified, wherein the data table to be classified includes the table name and field names;

[0031] The pre-classification module is used to classify the data table to be classified according to the table name, so as to establish a mapping relationship between the data table to be classified and the preset three-level directory, wherein the three-level directory has multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items.

[0032] The classification module is used to call the corresponding asset item in the fourth-level directory as the target asset item based on the mapping relationship, so as to determine the security level of the data table to be classified according to the security level of the target asset item with the highest similarity to the field name of the data table to be classified.

[0033] This application also provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the data asset classification and grading method.

[0034] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the data asset classification and grading method described above.

[0035] As described above, this application provides a data asset classification and grading method, system, device, and medium, which have the following beneficial effects.

[0036] This application first classifies and categorizes the data tables to be classified based on their names, determining the third-level directory to which the data tables belong. Then, it performs field matching based on the asset items corresponding to the fourth-level directories under the third-level directories, assigning the data tables to be classified to the corresponding fourth-level directories. Based on the security level of the target asset items in the fourth-level directories, it conducts a security assessment of the data tables to be classified, which facilitates subsequent management of data tables with different security levels, reduces manual intervention, and improves the efficiency of archiving and classification. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating a data asset classification and grading method in one embodiment of this application.

[0038] Figure 2 This is a schematic diagram of the overall architecture of the improved ERNIE-DPCNN model in one embodiment of this application.

[0039] Figure 3 This is a schematic diagram illustrating the process of matching field names with target asset items in one embodiment of this application.

[0040] Figure 4 A schematic diagram of the process for matching field names with target asset items in another embodiment of this application.

[0041] Figure 5 This is a block diagram of a data asset classification and grading system in one embodiment of this application.

[0042] Figure 6 This is a schematic diagram of the device in one embodiment of this application. Detailed Implementation

[0043] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0044] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0045] Explanation of technical terms:

[0046] ERNIE, Enhanced Language Representation with Informative Entities, employs a multi-stage knowledge masking strategy to enhance pre-trained language models. This strategy adds two more masking strategies (entity-level and phrase-level) to the basic level masking strategy used by BERT. After these improvements, compared to the BERT model, the ERNIE model can potentially learn longer semantic dependencies, resulting in greater generalization ability.

[0047] DPCNN, Deep Pyramid Convolutional Neural Networks for TextCategorization, takes text as input and passes it through a regionembedding layer containing three different convolutional feature extractors. The output can be considered as the embedding. Then, it passes through two layers of equal-length convolutions to provide a wider receptive field for subsequent feature extraction (improving the richness of the embedding). Finally, it repeatedly passes through a 1 / 2 pooling residual block. The 1 / 2 pooling continuously improves the semantics of word positions, fixing the feature maps. The residual network is introduced to address the vanishing and exploding gradient problems during training.

[0048] The Transformer model is a new generation network architecture proposed by Google in 2017, following the Convolutional Neural Network (CNN) model. It was initially used for Natural Language Processing (NLP), and now it also performs well in many downstream tasks in computer vision (such as image classification, object detection, image segmentation, etc.) and other fields.

[0049] BERT (Bidirectional Encoder Representations from Transformers) is a bidirectional encoder for Transformers designed to pre-train deep bidirectional representations from unlabeled text by conditional computation shared by both left and right contexts. Therefore, a pre-trained BERT model can be fine-tuned with only one additional output layer, enabling the generation of state-of-the-art models for a wide range of natural language processing tasks.

[0050] TransE (translating embedding) is a classic knowledge representation learning method that uses distributed representations to describe triples in a knowledge base. This type of representation avoids the construction of massive tree structures and can obtain semantic information through simple mathematical calculations, thus becoming the foundation of current representation learning.

[0051] Please see Figure 1 , Figure 1 This is a flowchart illustrating a data asset classification and grading method according to an embodiment of this application. This application provides a data asset classification and grading method, which includes the following steps:

[0052] Step S100: Obtain the data table to be classified, wherein the data table to be classified includes a table name and field names.

[0053] In one embodiment, the data table to be classified typically includes a table name and field names. For example, the table name can be "Member Funds," and the field names can include "Transaction Date," "Fund Account," "Member Code," "Speculative Margin," etc. The specific table name and field names can be adjusted according to actual application needs and are not limited here.

[0054] Step S110: Classify the data table to be classified according to its table name to establish a mapping relationship between the data table to be classified and a preset three-level directory, wherein the three-level directory has multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items.

[0055] In one embodiment, an asset catalog of data assets can be pre-built. The asset catalog may include a four-level catalog, with multiple asset items set in the fourth level catalog. Each asset item corresponds to a different security level. For example, the security levels can be divided into four levels, represented as follows: 4-Highly Sensitive: Information that, if leaked, illegally provided, or misused, is highly likely to cause significant damage to the economic interests, reputation, rights, and security of the relevant parties; 3-Moderately Sensitive: Information that, if leaked, illegally provided, or misused, is highly likely to cause some degree of damage to the economic interests, reputation, rights, and security of the relevant parties; 2-Suitable for Internal Disclosure (Low Sensitive): Information that is only used and shared internally, and if leaked, illegally provided, or misused, it may cause minor damage to the economic interests, reputation, rights, and security of the relevant parties; 1-Suitable for External Disclosure: Information that can be publicly shared. This is only an example of one security level classification rule. The specific security level classification can be adjusted according to actual application needs, and is not limited here.

[0056] In one embodiment, before classifying the data based on the table name of the data table to be classified, the following steps may be included:

[0057] Step S111: Obtain the sample data table, and concatenate the table name of the sample data table with the third-level directory as the input sample.

[0058] Step S112: Replace the phrases in the input samples with synonyms using a preset thesaurus, and expand the input samples to obtain an input sample set.

[0059] Step S113: Train a pre-classification model based on the input sample set, and classify the table names of the data table to be classified based on the pre-classification model.

[0060] Specifically, data tables related to public resources or historical data assets can be collected in advance as sample data tables. A multi-class classification task can be established, with the input data being the table names of the sample data tables, and processed as a supervised multi-class classification task. The format is shown in Table 1:

[0061] Table 1

[0062]

[0063] Furthermore, the data table name and the third-level directory are concatenated using the "|" symbol as input, with labels of 1 / 0 (1 for those with a mapping relationship and 0 for those without). The output consists of the three directories with the highest model scores. The format is shown in Table 2.

[0064] Table 2

[0065]

[0066] The input samples and labels constructed using the above methods can be used to train the model and obtain a pre-classification model. When the number of sample data is small, data augmentation methods can be used to expand the training data of the model and improve its generalization ability. For example, a training data entry "Exchanges that have obtained trading qualifications, authorized|Member Management / Institutional Management>Member Management>Professional Information" can be replaced with random synonyms to generate two new training data entries: "Exchanges that have obtained trading qualifications, authorized|Member Management / Department Management>Member Management>Professional Information" and "Exchanges that have obtained trading qualifications, authorized|Member Management / Institutional Management>Member Custody>Professional Information".

[0067] The specific model architecture can adopt conventional architectures such as support vector machines or neural networks. The choice and adjustment can be made according to the actual application requirements, and no restrictions are imposed here. The model training process is well known to those skilled in the art and will not be described in detail here.

[0068] After obtaining the pre-classification model, the table name of the data table to be classified can be input into the pre-classification model to obtain the third-level directory to which the table name belongs, thereby establishing a mapping relationship between different data tables to be classified and their corresponding third-level directories.

[0069] In another embodiment, before classifying the data based on the table name of the data table to be classified, the following steps may also be included:

[0070] Step S114: Obtain data samples;

[0071] Step S115: Extract the entities and relationships between entities from the data sample to obtain a knowledge graph of the data asset. The knowledge graph consists of triples of entity-relationship-entity correspondence.

[0072] Step S116: Train the model based on the data sample and the knowledge graph to obtain a classification model, and classify the table names of the data table to be classified based on the classification model.

[0073] Specifically, a data asset knowledge graph can be constructed, which is a graphical representation method for describing and organizing data assets. It is an ontology-based data management approach designed to improve an organization's ability to manage and utilize data assets. Common applications include: data management, where knowledge graphs help organizations classify, standardize, and integrate data, thereby improving data reusability and management efficiency; intelligent search, where knowledge graphs can quickly locate and find relevant entities, improving the accuracy and speed of search results; and the ability to uncover relationships and characteristics between entities, providing users with more intelligent recommendation services.

[0074] By using a knowledge graph to link and merge similar entities from different tables within the same table, and through a "symbolization" process, data fusion and collision mining are achieved. This completes a series of processes, including the extraction, fusion, reasoning, and accumulation of knowledge contained in industry data, thereby breaking down cognitive barriers and forming industry intelligence. Presenting governance results in the form of a graph allows clients to easily understand the process and outcomes, truly delivering the value of governance and helping them quickly and clearly utilize their managed data assets for in-depth data asset operation.

[0075] Before constructing the knowledge graph, the data samples can be preprocessed. Preprocessing includes defining nine entity nodes: basic assets, internal assets, external assets, indicator assets, category descriptions, departments, table names, field names, and reference standards. It also involves designing relationships between five entities: "Internal Asset - Internal / External Asset Mapping - External Asset," "Data Asset - Department Affiliation - Department," "Data Asset - Asset Affiliation - Category Description," "Field Name - Field Affiliation - Table Name," and "Field Name - Field Asset Item Mapping - Data Asset." Finally, the knowledge graph data is stored in a graph database in the form of triples.

[0076] In one embodiment, model training based on the data samples and the knowledge graph includes:

[0077] Convert the data samples into word vectors;

[0078] The triples in the knowledge graph are converted into embedded vector representations of entities and relations using a preset vector transformation model, thus obtaining the embedded vectors.

[0079] The word vectors and the embedding vectors are concatenated and then input into a preset initial classification model to establish a mapping relationship between data samples and different asset items, thereby obtaining the classification model.

[0080] In one embodiment, the initial classification model includes: multiple feature extraction units and a fully connected layer. The input of the initial classification model passes through multiple feature extraction units in sequence, and then passes through the fully connected layer for classification prediction to obtain the probability of different classification categories. The feature extraction unit includes multiple convolutional layers of equal length and a pooling layer.

[0081] Specifically, to build a more efficient training model suitable for Chinese text classification, this paper integrates ERNIE with an improved DPCNN and incorporates structured information using knowledge graph embedding to establish an improved ERNIE-DPCNN model. This model leverages the advantages of deep unsupervised pre-trained language models and Transformer long-distance feature extraction to better capture long-distance dependency information in text and effectively achieve feature extraction. On the other hand, it utilizes the deep pyramid neural network structure and weight sharing characteristics to fine-tune model parameters, which can greatly shorten the training time of the model and effectively avoid the gradient vanishing problem in deep neural networks.

[0082] Please see Figure 2 , Figure 2 This is a schematic diagram of the overall architecture of the improved ERNIE-DPCNN model in one embodiment of this application. The specific model construction process can be described as follows:

[0083] (I) After preprocessing the dataset, the input text is formed, denoted as W = (W1, W2, ..., W...). n ), where W i (i = 1, 2, 3, ..., n) represents the i-th text. We treat entities and relations in the knowledge graph as nodes, denoted as G = *N1, N2, ..., N n ).

[0084] (II) Each W i The input text W is converted into corresponding ERNIE word vectors by the ERNIE pre-training layer. That is, after passing through multiple bidirectional Transformer encoders, the input text W is represented by sequence features, and the output text X is generated. i =(X 1i ,X 2i ,...,X ji ), where X ji This represents the word vector of the j-th word in the i-th sentence. The word vectors in each sentence are concatenated using the concatenation operator to obtain the word vector matrix. At the same time, the TransE model is used to learn the representations of relationships and entities in the knowledge graph, and then the learned node vectors and word vectors are concatenated.

[0085] (III) The concatenated vector feature representation matrix X is used as input to the improved DPCNN model, and new feature representations are generated using convolutional layers of equal length:

[0086] E = f(KX + b), H = E + X

[0087] Where: H is the network output; K is the convolution kernel with size h; b is the partial convolution kernel.

[0088] The difference; f(x) = max(0,x) represents the ReLU nonlinear activation function.

[0089] (IV) After fixing the feature map size, E is max-pooled according to the improved DPCNN model, and the predicted classification probability is output through the normalized Softmax function. The sum of the predicted label probabilities of each class is 1, as shown in the formula:

[0090]

[0091]

[0092] Finally, the probability matrix P = (σ1(H), σ2(H), ..., σ2(H)) is obtained for each label class. n (H)), which is the classification label corresponding to the data sample: label 1, label 2, ..., label n.

[0093] After obtaining the classification model based on the above steps, a mapping relationship between the data table to be classified and the three-level directory can be established based on the classification model.

[0094] The foregoing embodiments provide two ways to establish the mapping relationship between the pending level data table and the third-level directory. In practical applications, one of them can be selected according to the needs, and there is no restriction here.

[0095] In addition to the mapping relationship established in the above embodiments, a thesaurus, prefix, and suffix table can also be established based on the naming rules and relationships between the table name and the existing data asset directory, so that some data tables with relatively standardized names can be directly mapped to the third- or fourth-level directories.

[0096] Step S120: Based on the mapping relationship, call the corresponding asset item in the fourth-level directory as the target asset item, so as to determine the security level of the data table to be classified according to the security level of the target asset item with the highest similarity to the field name of the data table to be classified.

[0097] In one embodiment, multiple fourth-level directories can typically be mounted under a third-level directory, and multiple asset items can be set in each fourth-level directory. After determining the mapping relationship between the data table to be classified and the third-level directory, the fourth-level directories mounted under the corresponding third-level directory can be determined based on the mapping relationship, and then the asset items under the fourth-level directory can be extracted as target asset items.

[0098] In one embodiment, before determining the security level of the data table to be classified based on the security level of the asset item with the highest similarity to the field names of the data table to be classified, the following steps are included:

[0099] Convert the field names of the data table to be classified into a first vector;

[0100] Convert the target asset item into a second vector;

[0101] Calculate the similarity between the first vector and the second vector to determine the target asset item with the highest similarity to the field name.

[0102] Specifically, please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the process of matching field names with target asset items in one embodiment of this application. Pre-trained models such as word2vec / fasttext / BERT can be used to obtain vector representations of field names and asset items. Then, the cosine similarity between vector 1 (i.e., the second vector) and vector 2 (i.e., the first vector) is calculated using a cosine similarity calculation method. If the similarity is greater than a set threshold, a mapping relationship is considered to exist between the field name and the target asset item.

[0103] In another embodiment, before determining the security level of the data table to be classified based on the security level of the highest asset item with the highest similarity to the field names of the data table to be classified, the method further includes:

[0104] The field names of the data table to be classified are concatenated with the target asset item to obtain a concatenation vector;

[0105] The concatenated vector is input into a pre-trained matching model to determine the target asset item with the highest similarity to the field name.

[0106] Specifically, please refer to Figure 4 , Figure 4 This is a flowchart illustrating the process of matching field names with target asset items in another embodiment of this application. The field value and asset item name are concatenated using "|", with labels of 1 / 0, and trained as a binary classification task. During prediction, a sigmoid function is added to the model's output value for normalization, and a threshold is set. A match is considered successful if the value exceeds this threshold. The deep learning model in the diagram is the pre-trained matching model. The deep learning model can employ deep neural networks or other network architectures; the specific deep learning network can be selected based on actual application requirements, and no restrictions are imposed here.

[0107] In one embodiment, all data tables are divided into a certain third-level directory. The asset items corresponding to the fourth-level directories under all third-level directories are matched with the field names in the data tables. The fourth-level directory with the highest matching rate is used as the final physical mapping result. If the matching rate is lower than a certain threshold, it is recommended to manually add new fourth-level directories to better improve the data asset directory.

[0108] After completing the directory matching, we can determine the security level of the data assets to be predicted by the security level of the four-level directory, and ensure that it is consistent with the level of the directory in which it is located, which facilitates management.

[0109] Based on the above technical solutions, when the directory matching is at a high confidence level, the security classification accuracy and the fourth-level directory mapping accuracy are both 87%; when the directory matching is at a low confidence level, the security classification model accuracy is 85%, so the overall accuracy of the security classification task is 86%. Using artificial intelligence to classify data asset security with high accuracy can save costs in various information and data management tasks, improve work efficiency, protect and standardize data resources, and enable the enterprise's internal data management platform to possess a certain degree of intelligent capability.

[0110] Please see Figure 5 This embodiment provides a data asset classification and grading system for executing the data asset classification and grading method described in the foregoing method embodiments. Since the technical principles of the system embodiment are similar to those of the foregoing method embodiments, the same technical details will not be repeated.

[0111] In one embodiment, a data asset classification and grading system includes: a data acquisition module 10, used to acquire a data table to be graded, wherein the data table to be graded includes a table name and field names; a pre-classification module 11, used to classify the data table to be graded according to the table name to establish a mapping relationship between the data table to be graded and a preset three-level directory, wherein the three-level directory has multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items; and a grading module 12, used to call the asset items in the corresponding four-level directories as target asset items based on the mapping relationship, so as to determine the security level of the data table to be graded according to the security level of the target asset item with the highest similarity to the field names of the data table to be graded.

[0112] This application also provides a data asset classification and grading device, which may include: one or more processors; and one or more machine-readable media storing instructions thereon, which, when executed by the one or more processors, cause the device to perform... Figure 1The method described herein. In practical applications, the device can function as a terminal device or a server. Examples of terminal devices include: smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptops, in-vehicle computers, desktop computers, set-top boxes, smart TVs, wearable devices, etc. This application does not limit the specific devices described.

[0113] This application also provides a machine-readable medium storing one or more modules (programs) that, when applied to a device, enable the device to execute embodiments of this application. Figure 1 The instructions for the steps included in the classification and grading method for data assets. Machine-readable media can be any usable medium that a computer can store, or a data storage device such as a server or data center that integrates one or more usable media. The usable medium can be magnetic media (such as floppy disks, hard disks, and magnetic tapes), optical media (such as DVDs), or semiconductor media (such as solid-state drives (SSDs)).

[0114] See Figure 6 This embodiment provides a device 80, which can be a desktop computer, a portable computer, a smartphone, or other devices. Specifically, the device 80 includes at least a memory 82 and a processor 83 connected via a bus 81. The memory 82 stores a computer program, and the processor 83 executes the computer program stored in the memory 82 to perform all or part of the steps in the aforementioned method embodiments.

[0115] The system bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0116] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0117] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for classifying and grading data assets, characterized in that, include: Obtain the data table to be classified, wherein the data table to be classified includes the table name and field names; The data tables to be classified are classified according to their table names to establish a mapping relationship between the data tables to be classified and the preset three-level directories. The three-level directories are mounted with multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items. Before classifying the data table to be classified according to its table name, the method further includes: obtaining data samples; extracting entities and relationships between entities from the data samples to obtain a knowledge graph of the data assets, wherein the knowledge graph consists of triples of entity-relationship-entity correspondence; training a model based on the data samples and the knowledge graph to obtain a classification model, and classifying the table name of the data table to be classified based on the classification model. Based on the mapping relationship, the corresponding asset item in the fourth-level directory is called as the target asset item, so as to determine the security level of the data table to be classified according to the security level of the target asset item with the highest similarity to the field name of the data table to be classified.

2. The data asset classification and grading method according to claim 1, characterized in that, Model training based on the data samples and the knowledge graph includes: Convert the data samples into word vectors; The triples in the knowledge graph are converted into embedded vector representations of entities and relations using a preset vector transformation model, thus obtaining the embedded vectors. The word vectors and the embedding vectors are concatenated and then input into a preset initial classification model to establish a mapping relationship between data samples and different asset items, thereby obtaining the classification model.

3. The data asset classification and grading method according to claim 2, characterized in that, The initial classification model includes multiple feature extraction units and a fully connected layer. The input of the initial classification model passes through multiple feature extraction units in sequence, and then passes through the fully connected layer for classification prediction to obtain the probability of different classification categories. The feature extraction unit includes multiple convolutional layers of equal length and a pooling layer.

4. The data asset classification and grading method according to claim 1, characterized in that, Before determining the security level of the data table to be classified based on the security level of the target asset item with the highest similarity to the field names of the data table to be classified, the following steps are included: Convert the field names of the data table to be classified into a first vector; Convert the target asset item into a second vector; Calculate the similarity between the first vector and the second vector to determine the target asset item with the highest similarity to the field name.

5. The data asset classification and grading method according to claim 1, characterized in that, Before determining the security level of the data table to be classified based on the security level of the target asset item with the highest similarity to the field names of the data table to be classified, the process also includes: The field names of the data table to be classified are concatenated with the target asset item to obtain a concatenation vector; The concatenated vector is input into a pre-trained matching model to determine the target asset item with the highest similarity to the field name.

6. The data asset classification and grading method according to claim 1, characterized in that, Before classifying the data based on the table name of the data table to be classified, the following steps are also included: Obtain the sample data table, and concatenate the table name of the sample data table with the third-level directory to use as the input sample; By replacing the phrases in the input samples with synonyms using a preset thesaurus, and then expanding the input samples to obtain an input sample set; A pre-classification model is trained based on the input sample set to classify the table names of the data table to be classified based on the pre-classification model.

7. A data asset classification and grading system, characterized in that, include: The data acquisition module is used to acquire the data table to be classified, wherein the data table to be classified includes the table name and field names; The pre-classification module is used to classify the data table to be classified according to the table name, so as to establish a mapping relationship between the data table to be classified and the preset three-level directory, wherein the three-level directory has multiple four-level directories with preset security levels, and each four-level directory includes multiple asset items. Before classifying the data table to be classified according to its table name, the method further includes: obtaining data samples; extracting entities and relationships between entities from the data samples to obtain a knowledge graph of the data assets, wherein the knowledge graph consists of triples of entity-relationship-entity correspondence; training a model based on the data samples and the knowledge graph to obtain a classification model, and classifying the table name of the data table to be classified based on the classification model. The classification module is used to call the corresponding asset item in the fourth-level directory as the target asset item based on the mapping relationship, so as to determine the security level of the data table to be classified according to the security level of the target asset item with the highest similarity to the field name of the data table to be classified.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the steps of the data asset classification and grading method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the data asset classification and grading method according to any one of claims 1 to 6.

Images