CN111488400A - Data classification method, apparatus and computer readable storage medium - Google Patents

Abstract

The present disclosure relates to a data classification method, an apparatus and a computer-readable storage medium, and relates to the field of computer technology. The method includes: generating a first to-be-processed graph according to each part of the data included in the data to be processed and the relationship between each part of the data, the first to-be-processed graph including nodes corresponding to each part of the data and according to the relationship between each part of the data The edge between each node determined by the relationship; according to the extracted feature vector of each part of the data and the relationship between each part of the data, use the machine learning model to determine the associated feature vector of each node in the first to-be-processed graph and each node. According to the associated feature vector of each node and the clustering result of each node, use the classifier to classify the data to be processed. The technical solutions of the present disclosure can improve the accuracy of data classification.

Description

Data classification method, apparatus and computer readable storage medium

technical field

The present disclosure relates to the field of computer technology, and in particular, to a data classification method, a data classification apparatus, and a computer-readable storage medium.

Background technique

MIL (Multiple Instance Learning) is a weakly supervised method that can be used to deal with weakly labeled data. In MIL, each data sample is a "bag", and a "bag" contains multiple "instances". A "package" has a category label, while an "example" has no category label.

Many application scenarios of machine learning tasks can be modeled as MIL models with such properties. For example, a comment in an e-commerce platform can contain multiple sentences, but a comment has only one annotation result; an image can contain multiple regions, but an image has only one classification result about the target.

In the related art, the degree of similarity between the packets is used for classification, or the packets are mapped into a vector, and then the classification is performed using a single-instance method.

SUMMARY OF THE INVENTION

The inventors of the present disclosure found that the above-mentioned related art has the following problems: the correlation of each example in a package is not considered, resulting in low accuracy of data classification.

In view of this, the present disclosure proposes a data classification technical solution, which can improve the accuracy of data classification.

According to some embodiments of the present disclosure, a data classification method is provided, including: generating a first to-be-processed graph according to each part of data included in the data to be processed and the relationship between the various parts of the data, the first to-be-processed map The processing graph includes the nodes corresponding to the data of each part and the edges between the nodes determined according to the relationship between the data of each part; according to the extracted feature vector of the data of each part and the data of each part The relationship between the two nodes, the machine learning model is used to determine the associated feature vectors of the corresponding nodes in the first to-be-processed graph and the clustering results of the nodes; according to the associated feature vectors of the nodes and the Clustering results, using a classifier to classify the data to be processed.

In some embodiments, using a classifier to classify the data to be processed according to the feature vectors of the nodes and the clustering results of the nodes includes: according to the associated feature vectors of the nodes and the clustering results class result, generate a second graph to be processed, the number of nodes in the second graph to be processed is determined according to the clustering result; according to the second graph to be processed, use the classifier to classify the data to be processed. sort.

In some embodiments, generating the second graph to be processed includes: performing pooling processing on the first graph to be processed according to the associated feature vector of each node and the clustering result; according to the result of the pooling processing, The second to-be-processed graph is generated.

In some embodiments, generating the second graph to be processed includes: generating a node feature vector matrix V of the second graph to be processed according to the formula V=S ^T Z, where S is the first graph in the clustering result A probability distribution matrix composed of the probability that each node in the to-be-processed graph belongs to each category, Z is an associated eigenvector matrix composed of associated eigenvectors of each node in the first to-be-processed graph; generated according to the formula A=S ^T A'S The edge matrix A of the second to-be-processed graph, A' is the edge matrix of the first to-be-processed graph determined according to the relationship between the parts of the data; the second to-be-processed graph is generated according to V and A .

In some embodiments, the machine learning model includes a first graph neural network model and a second graph neural network model; the machine learning model is used to determine the associated feature vector of each node in the first to-be-processed graph and the each node The clustering result includes: using the first graph neural network model to determine the associated feature vector of each node in the first graph to be processed; using the second graph neural network model to determine the clustering of each node result.

In some embodiments, the machine learning model further includes a third graph neural network model; according to the second graph to be processed, using the classifier to classify the data to be processed includes: using the third graph The graph neural network model determines the feature vector of each node in the second to-be-processed graph; and the classifier is used to classify the to-be-processed data according to the feature vector of each node in the second to-be-processed graph.

In some embodiments, according to the feature vector of each node in the second to-be-processed graph, using the classifier to classify the to-be-processed data includes: classifying the feature vector of each node in the second to-be-processed graph Perform maximum pooling or cascade processing to determine a vector to be processed; and use the classifier to classify the data to be processed according to the vector to be processed.

In some embodiments, the relationship between the parts of the data is determined according to the degree of association between the feature vectors of the parts of the data.

In some embodiments, the machine learning model and the classifier are trained using a cross-entropy loss function determined from a first classification result, a second classification result and a third classification result, the first classification result The classification result is the classification result determined by the classifier according to the associated feature vector of each node in the first to-be-processed graph, and the second classification result is based on the feature vector of each node in the second to-be-processed graph. The classification result determined by the classifier, and the third classification result is the classification result determined by the classifier according to the vector to be processed.

According to other embodiments of the present disclosure, a data classification apparatus is provided, comprising: a generating unit configured to generate a first to-be-processed graph according to each part of the data included in the data to be processed and the relationship between the various parts of the data, The first to-be-processed graph includes nodes corresponding to the parts of the data and edges between the nodes determined according to the relationship between the parts of the data; the determining unit is used to determine the parts according to the extracted parts The relationship between the feature vector of the data and the various parts of the data, using the machine learning model to determine the associated feature vector of the corresponding nodes in the first to-be-processed graph and the clustering results of the nodes; the classification unit, used for According to the associated feature vector of each node and the clustering result of each node, a classifier is used to classify the data to be processed.

According to yet other embodiments of the present disclosure, there is provided a data classification apparatus, comprising: a memory; and a processor coupled to the memory, the processor configured to execute, based on instructions stored in the memory apparatus, The data classification method in any one of the above embodiments.

According to still other embodiments of the present disclosure, there is provided a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the data classification method in any one of the foregoing embodiments.

In the above-mentioned embodiment, the data to be processed is modeled as a graph with each part of the data in the data to be processed as a node and the relationship between the parts of the data as an edge. Then, feature extraction and clustering are performed according to the associations between the nodes in the graph, and classification is performed based on the processing results. In this way, the correlation between each part of the data can be fully utilized for classification, thereby improving the accuracy of data classification.

Description of drawings

The accompanying drawings, which form a part of the specification, illustrate embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure.

The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, wherein:

FIG. 1 shows a flowchart of some embodiments of the data classification method of the present disclosure;

FIG. 2 shows a flowchart of some embodiments of step S30 of FIG. 1;

FIG. 3 shows a flowchart of some embodiments of step S320 of FIG. 2;

FIG. 4 shows a schematic diagram of some embodiments of the data classification method of the present disclosure;

FIG. 5 shows a schematic diagram of other embodiments of the data classification method of the present disclosure;

6 illustrates a block diagram of some embodiments of the data classification apparatus of the present disclosure;

7 shows a block diagram of other embodiments of the data classification apparatus of the present disclosure;

8 shows a block diagram of further embodiments of the data classification apparatus of the present disclosure.

Detailed ways

Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

Meanwhile, it should be understood that, for the convenience of description, the dimensions of various parts shown in the accompanying drawings are not drawn in an actual proportional relationship.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application or uses in any way.

Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the authorized description.

In all examples shown and discussed herein, any specific value should be construed as illustrative only and not as limiting. Accordingly, other examples of exemplary embodiments may have different values.

It should be noted that like numerals and letters refer to like items in the following figures, so once an item is defined in one figure, it does not require further discussion in subsequent figures.

FIG. 1 shows a flowchart of some embodiments of the data classification method of the present disclosure.

As shown in FIG. 1 , the method includes: step S10 , generating a first graph to be processed; step S20 , determining an associated feature vector and a clustering result; and step S30 , classifying the data to be processed.

In step S10, a first to-be-processed graph is generated according to each part of the data included in the to-be-processed data and the relationship between each part of the data. The first graph to be processed includes nodes corresponding to each part of the data and edges between each node determined according to the relationship between each part of the data. For example, the data to be processed is a package in the MIL model, and each part of the data is each example contained in the package.

In some embodiments, the data to be processed may be an image to be processed, and each area included in the image to be processed is each part of the data; the data to be processed may be a text to be processed, and each sentence included in the text to be processed is each part of the data.

In some embodiments, feature vectors of each part of the data may be extracted (eg, using image processing method, word vector method, etc.), and then the relationship between each part of the data is determined according to the degree of correlation between the feature vectors of each part of the data.

For example, the data to be processed is a package containing K examples, the feature vector set of the K examples is X=[x ₁ , x ₂ ,...x _k ,...x _K ], and x _k is the kth example The D-dimensional feature vector of , k is a positive integer less than K, and D is a positive integer. The distances between the eigenvectors can be calculated, the relationship parameters between the examples are determined according to the distances between the eigenvectors, and then the K×K-dimensional edge matrix A′ of the first graph to be processed is generated according to the following formula:

a _mn is the element of the mth row and nth column in A', dist(x _m , x _n ) is the distance between the feature vectors x _m and x _n , and m and n are positive integers less than or equal to K. η is the distance threshold, which is used to determine whether there is an edge between each node in the first graph to be processed (a _mn is 1, there is an edge, otherwise there is no edge). In the case of η=0, there is no edge in the first graph to be processed; in the case of η=+∞, the first graph to be processed is a complete graph.

For example, the first to-be-processed graph G ₁ =(A', V') can be established according to the edge matrix A', where V' is the feature vector matrix V'=X of the nodes in G ₁ .

In step S20, according to the extracted feature vector of each part of the data and the relationship between each part of the data, use the machine learning model to determine the associated feature vector of each node in the first to-be-processed graph and the clustering result of each node. For example, multiple clustering processes can be performed to obtain the desired clustering results.

In some embodiments, the machine learning model includes a first graph neural network model, a second graph neural network model. For example, the first graph neural network model is used to determine the associated feature vector of each node in the first graph to be processed; the second graph neural network model is used to determine the clustering result of each node.

In some embodiments, A' and V' may be input into the first graph neural network model for graph embedding processing, and the D'-dimensional associated feature vectors of each node in the first graph to be processed are determined to form associated feature vector matrices Z, D ' is a positive integer

It can be seen that the eigenvectors in V' are the eigenvectors of each example extracted independently, while the eigenvectors in Z are the eigenvectors of each example determined according to the correlation between the examples (relationship between the eigenvectors). That is to say, the feature vector in Z contains the relationship between the examples, and using Z to classify the data to be processed can improve the accuracy of the classification.

In some embodiments, A', V' may be input into the second graph neural network model for clustering processing, and the clustering result of each node in the first graph to be processed is determined. For example, the second graph neural network model divides each node in the first graph to be processed into C classes, where C is a positive integer greater than 1 set according to the actual situation. For example, each node can be clustered into two types, positive samples and negative samples for classification.

The softmax function can be used to process the output of the neural network model of the second graph to obtain the probability that each node in the first graph to be processed belongs to each category, and then obtain a probability distribution matrix S.

It can be seen that S contains the association relationship (clustering information) of each node. Using S to classify the data to be processed can improve the classification accuracy.

In step S30, the data to be processed is classified by a classifier according to the associated feature vector of each node and the clustering result of each node. For example, the classifier can be configured according to the MLP (Multi-Layer Perception) model.

In some embodiments, step S30 may be implemented by the embodiment in FIG. 2 .

FIG. 2 shows a flowchart of some embodiments of step S30 of FIG. 1 .

As shown in FIG. 2 , step S30 includes: step S310 , generating a second graph to be processed; and step S320 , classifying the data to be processed.

In step S310, a second graph to be processed is generated according to the associated feature vector of each node and the clustering result, and the number of nodes in the second graph to be processed is determined according to the clustering result.

In some embodiments, pooling is performed on the first graph to be processed according to the associated feature vector of each node and the clustering result. Then, according to the result of the pooling process, a second graph to be processed is generated. For isomorphic graphs, no matter how the nodes are arranged in the graph, the pooling process can ensure that the obtained processing results (the representation of the graph) are the same.

For example, the node eigenvector matrix V of the second graph to be processed is generated according to the formula V=S ^T Z, and the eigenvectors in it are D'dimension; the edge matrix A of the second graph to be processed is generated according to the formula A=S ^T A'S ; and then generate a second to-be-processed graph G ₂ =(A, V).

It can be seen that G ₂ is generated according to the association information (association between feature vectors, clustering results) among the examples, and the number of nodes in G ₂ is smaller than the number of nodes in G ₁ . In this way, the "big picture" that does not contain associated information in the feature vector of the node is converted into a "small picture" that includes the associated information in the feature vector of the node, thereby improving the accuracy of data classification.

In step S320, according to the second to-be-processed graph, the classifier is used to classify the data to be processed.

In some embodiments, the machine learning model further includes a third graph neural network model. The feature vector of each node in the second to-be-processed graph can be determined by using the third graph neural network model. That is, using the third graph neural network model to perform graph embedding processing on each node in the second graph to be processed.

For example, in some embodiments, if C is set to 1, there is only one node in G ₂ , and the graph embedding process can be regarded as a learned graph representation; _The joint processing completes the graph embedding processing of the nodes in G2.

In some embodiments, GNN (Graph Neural Network, Graph Neural Network) has two basic processes, ie, aggregation process and combination process. Aggregation processing is used to collect information from neighbor nodes of the current node, and joint processing is used to fuse the collected information to express the current node. Performing an aggregation process and a joint process is called a hop. The aggregation processing and joint processing of the first hop of GNN can be expressed by the following formulas respectively:

和

分别为第l跳的聚合处理结果和联合处理结果。

即为经过l跳学习得到的节点v的表达(如特征向量)。f _agg () and f _com () are the aggregation processing function and the joint processing function, respectively, N(v) represents the neighbor node of node v in the graph,

and

are the aggregate processing results and joint processing results of the lth hop, respectively.

That is, the expression of node v (such as feature vector) obtained after l-hop learning.

For example, according to the above description, the aggregation processing and joint processing of any GNN model in the above embodiments can be synthesized into the following formula:

The expression Vec _v of the node v can be obtained through the above formula, Vec _u is the representation of the neighbor node u, act() is the activation function (such as the LeakyReLu function, etc.), and MEAN() is the mean value function.

In some embodiments, step S320 may be implemented by the embodiment in FIG. 3 .

FIG. 3 shows a flowchart of some embodiments of step S320 of FIG. 2 .

As shown in FIG. 3 , step S320 includes: step S3210 , determining a vector to be processed; and step S3220 , classifying the data to be processed.

In step S3210, maximum pooling or cascade processing is performed on the feature vectors of each node in the second to-be-processed graph to determine the to-be-processed vector. For example, according to the number C of classes in the clustering result, it may be determined to perform maximum pooling processing or cascade processing to obtain a fixed-dimensional graph embedding result (vector to be processed).

In step S3220, the data to be processed is classified by a classifier according to the vector to be processed.

FIG. 4 shows a schematic diagram of some embodiments of the data classification method of the present disclosure.

As shown in FIG. 4 , the first to-be-processed map 41 can be generated according to the MIL model as the data to be processed (eg, an image with multiple regions, a text with multiple sentences, etc.). The to-be-processed graph 41 contains multiple nodes, corresponding to multiple examples of the MIL model (such as regions, sentences, etc.). Each node of the to-be-processed graph 41 has a feature vector, and the feature vector is the feature vector of each example in the MIL model (for example, it can be extracted by methods such as image processing and natural language processing). The edges in the to-be-processed graph 41 are the relationships between the nodes (for example, it can be determined according to the feature vectors of the nodes).

The nodes in the to-be-processed figure 41 may be processed by graph embedding using the first graph neural network model in any of the above embodiments, so as to obtain the associated feature vectors of the nodes in the to-be-processed figure 41 .

The nodes in the to-be-processed FIG. 41 can be clustered by the second graph neural network model in any of the above embodiments.

In some embodiments, the second graph neural network model can be trained unsupervised to be able to distinguish between positive samples (positively labeled examples) and negative samples (negatively labeled examples) in the MIL model, eg, through a clustering process The nodes in Figure 41 to be processed are automatically divided into classes 421 and 422, the nodes in class 421 are positive samples, and the nodes in class 422 are negative samples.

The to-be-processed graph 42 can be generated according to the associated feature vector and the clustering result. For example, the to-be-processed graph 42 has two nodes, which may correspond to class 421 and class 422 respectively.

The to-be-processed graph 42 is subjected to graph embedding processing and pooling processing, and a to-be-processed vector 44 can be obtained. The to-be-processed vector 44 can be viewed as a graph with only one node.

The above embodiment converts the MIL model into a "big picture", then converts the "big picture" into a "small picture", and further converts the "small picture" into a fixed-dimensional "vector" with only one node (the vector to be processed). ), and process the "vector" to get the classification result. In this way, in the classification process, not only the correlation between the examples is fully considered, but also the dimensionality reduction process is carried out, thereby improving the classification accuracy and efficiency.

FIG. 5 shows a schematic diagram of other embodiments of the data classification method of the present disclosure.

As shown in Figure 5, the machine learning models include GNN 51, GNN 52, GNN 53, and MLP 54.

The GNN 51 is used to perform graph embedding processing on the nodes in the first graph to be processed, so as to obtain the associated feature vector of each node. The GNN 52 is configured to perform graph clustering processing on the nodes in the first graph to be processed, so as to obtain the clustering results of each node. According to the output results of GNN 51 and GNN 52, the second graph to be processed can be further acquired.

In some embodiments, in the training phase, the first classification result determined by the classifier may be used according to the associated feature vector (output of the GNN 51 ) of each node in the first graph to be processed.

In some embodiments, in the training phase, the second classification result determined by the classifier may be used according to the feature vector of each node in the second graph to be processed.

The GNN 53 is used to perform graph embedding processing on the nodes in the second graph to be processed, so as to obtain feature vectors of each node. The second to-be-processed graph may also be further processed to obtain a to-be-processed vector. The MLP 54 is used to classify the vector to be processed to obtain a third classification result.

In some embodiments, in the training phase, a cross-entropy loss function may be determined according to the first classification result, the second classification result and the third classification result, and the machine learning model is trained by using the cross-entropy loss function. In this way, processing results of different depths in the machine learning model can be used for training, thereby improving the training efficiency and accuracy of the machine learning model.

In the above-mentioned embodiment, the data to be processed is modeled as a graph with each part of the data in the data to be processed as a node and the relationship between the parts of the data as an edge. Then, feature extraction and clustering are performed according to the associations between the nodes in the graph, and classification is performed based on the processing results. In this way, the correlation between each part of the data can be fully utilized for classification, thereby improving the accuracy of data classification.

Figure 6 shows a block diagram of some embodiments of the data classification apparatus of the present disclosure.

As shown in FIG. 6 , the data classification apparatus 6 includes a generation unit 61 , a determination unit 62 and a classification unit 63 .

The generating unit 61 generates a first to-be-processed graph according to the relationship between each part of the data included in the to-be-processed data. The first graph to be processed includes nodes corresponding to each part of the data and edges between each node determined according to the relationship between each part of the data. For example, the relationship between the parts of the data is determined according to the degree of association between the feature vectors of the parts of the data.

In some embodiments, the generating unit 61 generates the second graph to be processed according to the associated feature vector of each node and the clustering result. The generating unit 61 may perform pooling processing on the first graph to be processed according to the associated feature vector of each node and the clustering result. The generating unit 61 may generate the second graph to be processed according to the result of the pooling process.

For example, the generating unit 61 generates the node feature vector matrix V of the second graph to be processed according to the formula V= ^ST Z. S is a probability distribution matrix composed of probabilities that each node in the first to-be-processed graph belongs to each category in the clustering result, and Z is an associated eigenvector matrix composed of associated eigenvectors of each node in the first to-be-processed graph. The generating unit 61 generates the edge matrix A of the second graph to be processed according to the formula A= ^ST A'S. A' is the edge matrix of the first graph to be processed determined according to the relationship between each part of the data. The generating unit 61 generates a second map to be processed according to V and A.

The determining unit 62 uses the machine learning model to determine the associated feature vectors of the corresponding nodes in the first to-be-processed graph and the clustering results of the nodes according to the extracted feature vectors of each part of the data and the relationship between each part of the data.

In some embodiments, the machine learning model includes a first graph neural network model, a second graph neural network model. The determining unit 62 uses the first graph neural network model to determine the associated feature vector of each node in the first graph to be processed. The determining unit 62 uses the second graph neural network model to determine the clustering result of each node.

In some embodiments, the machine learning model further includes a third graph neural network model. The determining unit 62 uses the neural network model of the third graph to determine the feature vector of each node in the second graph to be processed.

In some embodiments, the determining unit 62 performs maximum pooling processing or cascade processing on the feature vectors of each node in the second to-be-processed graph to determine the to-be-processed vector.

In some embodiments, machine learning models and classifiers are trained using a cross-entropy loss function. For example, the cross-entropy loss function is determined according to the first classification result, the second classification result, and the third classification result. The first classification result is the classification result determined by the classifier according to the associated feature vector of each node in the first to-be-processed graph. The second classification result is the classification result determined by the classifier according to the feature vector of each node in the second to-be-processed graph. The third classification result is the classification result determined by the classifier according to the vector to be processed.

The classification unit 63 uses a classifier to classify the data to be processed according to the associated feature vector of each node and the clustering result of each node.

In some embodiments, the number of nodes in the second graph to be processed is determined according to the clustering result. The classification unit 63 uses a classifier to classify the data to be processed according to the second to-be-processed map.

In some embodiments, the classification unit 63 uses a classifier to classify the data to be processed according to the feature vector of each node in the second to-be-processed graph.

In some embodiments, the classification unit 63 uses a classifier to classify the data to be processed according to the vector to be processed.

In the above-mentioned embodiment, the data to be processed is modeled as a graph with each part of the data in the data to be processed as a node and the relationship between the parts of the data as an edge. Then, feature extraction and clustering are performed according to the associations between the nodes in the graph, and classification is performed based on the processing results. In this way, the correlation between each part of the data can be fully utilized for classification, thereby improving the accuracy of data classification.

FIG. 7 shows a block diagram of other embodiments of the data classification apparatus of the present disclosure.

As shown in FIG. 7 , the data classification apparatus 7 of this embodiment includes: a memory 71 and a processor 72 coupled to the memory 71 , and the processor 72 is configured to execute any of the instructions in the present disclosure based on the instructions stored in the memory 71 A data classification method in one embodiment.

Wherein, the memory 71 may include, for example, a system memory, a fixed non-volatile storage medium, and the like. The system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader), a database, and other programs.

8 shows a block diagram of further embodiments of the data classification apparatus of the present disclosure.

As shown in FIG. 8 , the data classification apparatus 8 of this embodiment includes: a memory 810 and a processor 820 coupled to the memory 810 , and the processor 820 is configured to execute any one of the foregoing implementations based on instructions stored in the memory 810 The data classification method in the example.

Memory 810 may include, for example, system memory, fixed non-volatile storage media, and the like. The system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader), and other programs.

The data classification device 8 may further include an input and output interface 830, a network interface 840, a storage interface 850, and the like. These interfaces 830 , 840 , 850 and the memory 810 and the processor 820 may be connected, for example, through a bus 860 . The input and output interface 830 provides a connection interface for input and output devices such as a display, a mouse, a keyboard, and a touch screen. Network interface 840 provides a connection interface for various networked devices. The storage interface 850 provides a connection interface for external storage devices such as SD cards and U disks.

As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein .

So far, the data classification method, the data classification apparatus, and the computer-readable storage medium according to the present disclosure have been described in detail. Some details that are well known in the art are not described in order to avoid obscuring the concept of the present disclosure. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

The methods and systems of the present disclosure may be implemented in many ways. For example, the methods and systems of the present disclosure may be implemented in software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Furthermore, in some embodiments, the present disclosure can also be implemented as programs recorded in a recording medium, the programs including machine-readable instructions for implementing methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

While some specific embodiments of the present disclosure have been described in detail by way of examples, those skilled in the art will appreciate that the above examples are provided for illustration only, and are not intended to limit the scope of the present disclosure. Those skilled in the art will appreciate that modifications may be made to the above embodiments without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims (12)

1. A data classification method, comprising: A first graph to be processed is generated according to each part of the data included in the data to be processed and the relationship between the parts of the data, and the first graph to be processed includes nodes corresponding to the parts of the data and Edges between nodes determined by the relationship between part of the data; According to the extracted feature vector of each part of the data and the relationship between the each part of the data, use the machine learning model to determine the associated feature vector of each node in the first to-be-processed graph and the clustering of each node result; According to the associated feature vector of each node and the clustering result of each node, a classifier is used to classify the data to be processed. 2. The data classification method according to claim 1, wherein, according to the feature vector of each node and the clustering result of each node, using a classifier to classify the data to be processed comprises: According to the associated feature vector of each node and the clustering result, a second graph to be processed is generated, and the number of nodes in the second graph to be processed is determined according to the clustering result; According to the second to-be-processed map, the classifier is used to classify the to-be-processed data. 3. The data classification method according to claim 2, wherein generating the second graph to be processed comprises: performing pooling processing on the first graph to be processed according to the associated feature vector of each node and the clustering result; According to the result of the pooling process, the second graph to be processed is generated. 4. The data classification method according to claim 3, wherein generating the second graph to be processed comprises: The node feature vector matrix V of the second graph to be processed is generated according to the formula V=S ^T Z, and S is the probability distribution matrix composed of the probabilities that each node in the first to-be-processed graph belongs to each category in the clustering result , Z is the associated eigenvector matrix formed by the associated eigenvectors of each node in the first to-be-processed graph; Generate the edge matrix A of the second graph to be processed according to the formula A= ^ST A'S, where A' is the edge matrix of the first graph to be processed determined according to the relationship between the parts of the data; The second pending map is generated from V and A. 5. The data classification method according to claim 2, wherein, The machine learning model includes a first graph neural network model and a second graph neural network model; Using a machine learning model to determine the associated feature vector of each node in the first to-be-processed graph and the clustering result of each node includes: Using the first graph neural network model, determine the associated feature vector of each node in the first graph to be processed; Using the second graph neural network model, the clustering result of each node is determined. 6. The data classification method according to claim 2, wherein, The machine learning model also includes a third graph neural network model; According to the second to-be-processed map, using the classifier to classify the to-be-processed data includes: Using the third graph neural network model, determine the feature vector of each node in the second graph to be processed; According to the feature vector of each node in the second to-be-processed graph, the classifier is used to classify the to-be-processed data. 7. The data classification method according to claim 6, wherein, according to the feature vector of each node in the second to-be-processed graph, using the classifier to classify the to-be-processed data comprises: Perform maximum pooling or cascade processing on the feature vectors of each node in the second to-be-processed graph to determine the to-be-processed vector; According to the to-be-processed vector, the classifier is used to classify the to-be-processed data. 8. The data classification method according to any one of claims 1-7, wherein, The relationship between the parts of the data is determined according to the degree of association between the feature vectors of the parts of the data. 9. The data classification method according to claim 7, wherein, The machine learning model and the classifier are trained by using a cross-entropy loss function, the cross-entropy loss function is determined according to the first classification result, the second classification result and the third classification result, and the first classification result is based on the The associated feature vector of each node in the first graph to be processed is determined by using the classifier, and the second classification result is determined by using the classifier according to the feature vector of each node in the second graph to be processed. Classification result, the third classification result is a classification result determined by the classifier according to the vector to be processed. 10. A data classification device, comprising: A generating unit, configured to generate a first to-be-processed graph according to each part of the data included in the data to be processed and the relationship between the various parts of the data, and the first to-be-processed graph includes nodes corresponding to the each part of the data and the edges between the nodes determined according to the relationship between the parts of the data; The determining unit is configured to use a machine learning model to determine, according to the extracted feature vector of each part of the data and the relationship between the each part of the data, the associated feature vector of the corresponding nodes in the first to-be-processed graph and the Clustering results of each node; A classification unit, configured to use a classifier to classify the data to be processed according to the associated feature vector of each node and the clustering result of each node. 11. A data classification device, comprising: memory; and A processor coupled to the memory, the processor configured to perform the data classification method of any of claims 1-9 based on instructions stored in the memory device. 12. A computer-readable storage medium on which a computer program is stored, which implements the data classification method according to any one of claims 1-9 when the program is executed by a processor.

Images