CN114564516A - Business object classification method, device, equipment and storage medium - Google Patents
Business object classification method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114564516A CN114564516A CN202210203772.XA CN202210203772A CN114564516A CN 114564516 A CN114564516 A CN 114564516A CN 202210203772 A CN202210203772 A CN 202210203772A CN 114564516 A CN114564516 A CN 114564516A
- Authority
- CN
- China
- Prior art keywords
- node
- probability
- weight
- neural network
- business object
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000013528 artificial neural network Methods 0.000 claims abstract description 107
- 239000013598 vector Substances 0.000 claims abstract description 40
- 238000013507 mapping Methods 0.000 claims description 43
- 238000013145 classification model Methods 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 42
- 238000004590 computer program Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000002596 correlated effect Effects 0.000 claims description 6
- 230000000875 corresponding effect Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 2
- 238000012545 processing Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 230000010354 integration Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 238000007635 classification algorithm Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000000295 complement effect Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Fuzzy Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a method, a device, equipment and a storage medium for classifying business objects, wherein the method comprises the following steps: converting the business object into a graph neural network, wherein the business object has a plurality of characteristic data, and nodes in the graph neural network represent a plurality of characteristic vectors of the business object; calculating the degree of association between nodes in the graph neural network as the node degree of the nodes; executing the graph neural network to output a first probability that the business object belongs to a preset category; identifying a second probability that the business object belongs to a preset category by using the feature vector; for the same category, the first probability and the second probability are fused into a third probability according to the node degree; and determining the class to which the business object belongs according to the third probability. In the embodiment, the classification of the community node prediction by using the node degree is organically combined with the classification of the single node prediction based on the characteristic vector, so that the classification precision is improved.
Description
Technical Field
The present application relates to the field of computer processing technologies, and in particular, to a method, an apparatus, a device, and a storage medium for classifying business objects.
Background
In business scenes such as community mining and anomaly detection, business objects such as users, videos and audios are classified, and different classification algorithms have different advantages and disadvantages, so that a plurality of classification algorithms are often used for classification, and a plurality of classification results are integrated into a final classification result through an integration mechanism.
Currently, an average value method is mostly used in an integration mechanism, that is, a plurality of classification algorithms output a plurality of score vectors to a service object, the average value of all the score vectors is calculated to be a final score vector, and a category corresponding to the largest score vector is taken as a final classification result.
The average method has a single processing mode for each fractional vector, uses a uniform processing mode for different business objects, and has low classification precision.
Disclosure of Invention
The application provides a method, a device, equipment and a storage medium for classifying business objects, which aim to improve the precision of classification through an integration mechanism.
According to an aspect of the present application, a method for classifying business objects is provided, including:
converting a business object into a graph neural network, the business object having a plurality of feature vectors, nodes in the graph neural network representing the plurality of feature vectors of the business object;
calculating the association degree between the nodes in the graph neural network as the node degree of the nodes;
executing the graph neural network to output a first probability that the business object belongs to a preset category;
executing a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector;
for the same category, fusing the first probability and the second probability into a third probability according to the node degree;
and determining the class to which the business object belongs according to the third probability.
According to another aspect of the present application, there is provided a classification apparatus for business objects, including:
the graph neural network conversion module is used for converting a business object into a graph neural network, wherein the business object is provided with a plurality of eigenvectors, and nodes in the graph neural network represent the plurality of eigenvectors of the business object;
the node degree calculation module is used for calculating the association degree between the nodes in the graph neural network as the node degree of the nodes;
the first classification module is used for executing the graph neural network to output a first probability that the business object belongs to a preset class;
the second classification module is used for executing a preset classification model so as to identify a second probability that the business object belongs to a preset class by using the feature vector;
a probability fusion module, configured to fuse, according to the node degree, the first probability and the second probability into a third probability for the same category;
and the class determining module is used for determining the class to which the business object belongs according to the third probability.
According to another aspect of the present application, there is provided a classification device for a business object, the classification device for a business object comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program executable by the at least one processor, the computer program being executable by the at least one processor to enable the at least one processor to perform the method of classifying a business object as described in any of the embodiments of the present application.
According to another aspect of the present application, a computer-readable storage medium is provided, which stores a computer program for causing a processor to implement a classification method of a business object according to any one of the embodiments of the present application when the computer program is executed.
In the embodiment, the business object is converted into a graph neural network, the business object has a plurality of eigenvectors, and nodes in the graph neural network represent the plurality of eigenvectors of the business object; calculating the degree of association between nodes in the graph neural network as the node degree of the nodes; executing the graph neural network to output a first probability that the business object belongs to a preset class; executing a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector; for the same category, the first probability and the second probability are fused into a third probability according to the node degree; and determining the class to which the business object belongs according to the third probability. The graph neural network has advantages for the classification prediction of the business objects with strong correlation degree and has disadvantages for the classification prediction of the business objects with weak correlation degree, and the classification model has advantages for the classification prediction of the business objects with weak correlation degree, the business object classification prediction with strong association degree has disadvantages, the business object classification prediction and the business object classification prediction have complementary relation, the business object classification prediction and the business object classification prediction can ensure the comprehensiveness of classification effect, the influence of the association degree between the business objects represented by the node degree on the graph neural network and the classification model is considered, the classification result of the neural network of the graph and the classification result of the classification model can be fused according to the degree of association between business objects represented by the node degree, the node degree is different under different business scenes, the classification result of the fusion graph neural network and the classification result of the classification model can be flexibly adjusted, and the classification precision is greatly improved.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present application, nor do they limit the scope of the present application. Other features of the present application will become apparent from the following description.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flowchart of a classification method for business objects according to an embodiment of the present application;
FIG. 2 is an exemplary diagram of a graph neural network provided in accordance with an embodiment of the present application;
fig. 3 is a schematic structural diagram of a classification apparatus for business objects according to a second embodiment of the present application;
fig. 4 is a schematic structural diagram of a classification device for a business object, which implements the classification method for a business object according to the embodiment of the present application.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example one
Fig. 1 is a flowchart of a classification method for a business object according to an embodiment of the present disclosure, where this embodiment is applicable to a case where multiple classification results are integrated into a final classification result through an integration mechanism based on a node degree, and the method may be executed by a classification device for a business object, where the classification device for a business object may be implemented in a form of hardware and/or software, and the classification device for a business object may be configured in a classification device for a business object. As shown in fig. 1, the method includes:
In different business scenarios, there are different business objects, which are collections of data with business domain characteristics.
For example, for the user-oriented service field, the service object may be a user, for the news media field, the service object may be news data, and for the mobile communication field, the service object may be mobile communication data; for the field of Electronic Commerce (EC), the business object may be advertisement data, for the field of automatic driving, the business object may be a point cloud, and so on.
The business objects, while carrying business characteristics of different business domains, are still data in nature, e.g., textual information, image data, audio data, video data, and so forth.
For these service objects, a method for configuring the service can be invoked in advance according to specific service requirements in a service scene to extract features from the service objects, and the features are expressed in a vector form to form a plurality of feature vectors, that is, when the service objects are classified, the service objects have a plurality of feature vectors and belong to multidimensional vectors.
Illustratively, if the service is to identify whether the user (client) is a web crawler, the behavior of the user (client) accessing a URI (Uniform Resource Identifier) may be extracted as a feature vector, such as the number of accesses to various URIs, the time of accesses to various URIs, the density of accesses to various URIs, and so on.
In this embodiment, the business objects are converted into Graph Neural Networks (GNNs), which are a generalized Neural network based on Graph structure and a connection model, and capture the Graph dependency relationship through message passing between the nodes of the Graph. Unlike standard neural networks, a graph neural network retains a state that can represent information from its neighborhood with arbitrary depth.
Wherein a graph is a data structure that models a set of objects (nodes) and their relationships (edges).
In general, a graph neural network can be divided into five categories, which are: graph Convolution Networks (GCNs), Graph Attention Networks (Graph Attention Networks), Graph Autoencoders (Graph Autoencoders), Graph generation Networks (Graph generating Networks), and Graph spatio-temporal Networks (Graph Spatial-temporal Networks).
The inputs to the graph neural network include a graph G in which a set of nodes (Vertex, also called vertices) V ═ V, and a feature vector Feat1,v2,…,vpE, the set of edges E ═ E1,e2,…,eqOn any edge ei＝(vj,vk) Denotes eiConnecting node vjAnd node vkEach node viCorresponding to an n-dimensional feature vector Feati＝{f1,f2,…,fn}。
In this embodiment, a plurality of feature vectors of the business objects are input into a graph neural network, and a plurality of nodes and a plurality of edges are output, where the nodes in the graph neural network represent the plurality of feature vectors of the business objects, and the edges in the graph neural network represent the relationship between the business objects, that is, there is a certain correlation between the business objects having edges.
And 102, calculating the association degree between the nodes in the graph neural network as the node degree of the nodes.
In the graph neural network, there are certain correlations (also called correlations) between some nodes and other nodes, and these correlations appear in the form of the number of edges, weights, and the like.
The node degree of the node can be used for distinguishing the type of the node, and the type includes a community node and an isolated node, wherein the community node is also called a non-isolated node, is a node with a higher association degree with other nodes, and represents that the node degree of the node is greater than or equal to a preset association threshold, the isolated node is a node with a lower association degree with other nodes, even zero, and represents that the node degree of the node is greater than or equal to a preset association threshold, that is, the node degree of the community node is greater than the node degree of the isolated node.
Taking the number of edges as an example, the number of edges connected (also referred to as associated) by a node can be counted for each node in the graph neural network as the node degree of the node, wherein one edge connects (also referred to as associated) two nodes.
In this example, when the number of edges connected by a node is equal to zero, it indicates that the node does not have any associated edges with other nodes, and the node is an isolated node.
When the number of edges connected with a node is greater than or equal to one, the node is represented as a community node when the node has any associated edges with other nodes.
Then, the association threshold may be set to one, the isolated node may be a node whose node degree is less than one (association threshold), and the community node may be a node whose node degree is greater than or equal to one (association threshold).
In one statistical approach, in the graph neural network, let the set of nodes V ═ V1,v2,…,vpThe set of edges E ═ E } ═ E1,e2,…,eqH, for each node v in the graph neural networki，viE.g. V, to node ViDegree of initialization node is zero
Querying edges e in graph neural networksi＝(vj,vk)，eiE denotes EiConnecting node vjAnd node vk，vj、vk∈V。
For each edge in the graph neural network, the node degrees of two nodes connected to the edge are accumulated by one, that is,
there are 9 nodes in total, node v, in the graph neural network shown in FIG. 21Has a node degree of 5, node v2Has a node degree of 2, node v3Has a node degree of 2, node v4Has a node degree of 0, a node v5Has a node degree of 2, node v6Has a node degree of 5, node v7Has a node degree of 4, node v8Has a node degree of 3, node v9Has a node degree of 1, node v1、v2、v3、v5、v6、v7、v8、v9Are all community nodes, node v4Are isolated nodes.
Of course, the above manner of calculating the node degrees is only an example, and when the embodiment is implemented, other manners of calculating the node degrees may be set according to actual situations, for example, summing weights of all edges associated with the node degrees, which are used as the node degrees of the node, and the like, which is not limited in this embodiment of the application. In addition, besides the above-mentioned manner of calculating the node degree, a person skilled in the art may also adopt other manners of calculating the node degree according to actual needs, and this embodiment also does not limit this.
And 103, executing the graph neural network to output a first probability that the business object belongs to a preset class.
In this embodiment, a complete graph neural network may be trained in advance in an end-to-end manner according to specific service requirements in a service scenario.
In classifying business objects, a graph neural network is executed, wherein the graph neural network takes an underlying graph as a computational graph and learns neural network primitives by transferring, converting and aggregating characteristics of nodes on the whole graph to generate embedded vectors (i.e., feature vectors) of single nodes, and the generated embedded vectors can be used as input of a micro-predictable layer for node classification.
In the present embodiment, v is calculated for each nodeiThe graph neural network outputs a vector S with m dimensionsi＝(s1,s2,…,sm)，Wherein s istRepresents the node viBelonging to a predetermined class CtIs denoted as the first probability, where t ∈ m, m is a positive integer.
And 104, executing a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector.
In this embodiment, one or more classification models may be trained in advance according to specific service requirements in a service scenario, where the classification model is a model independent of a graph neural network and is a model for classifying based on feature vectors, that is, a service object including only feature vectors is input, and a probability (also referred to as a score) that the service object belongs to a preset class is output.
Further, the classification model may apply a Machine learning algorithm, for example, an SVM (support vector machines), a LightGBM (Light Gradient Boosting Machine), or the like, and the classification model may apply a deep learning algorithm, for example, CNN (Convolutional Neural Network), RNN (current Neural Network), or the like.
For the deep learning algorithm, the structure of the classification model is not limited to the artificially designed Neural network, and may also be the Neural network optimized by the model quantization method, the Neural network searched for the specific service requirement in the service scene by the NAS (Neural Architecture Search) method, and the like, which is not limited in this embodiment.
In the present embodiment, v is calculated for each nodeiInputting n-dimensional feature vector Feat as business objecti＝{f1,f2,…,fnOutputting a m-dimensional vector S by each classification modeli＝(s1,s2,…,sm) Wherein s istRepresents the node viBelonging to a predetermined class CtIs denoted as a second probability, wherein t ∈ m, and m is a positive integer.
And 105, fusing the first probability and the second probability into a third probability according to the node degree aiming at the same category.
For the same business object, the graph neural network predicts a first probability that the business object belongs to a plurality of (two or more) classes, and at the same time, each classification model predicts a second probability that the business object belongs to a plurality of (two or more) classes, the classes into which the graph neural network divides the business object are the same as the classes into which the classification network divides the business object, so that for the same class, the same business object has at least two probabilities (i.e., the first probability, the one or more second probabilities) that belong to the class.
The graph neural network has advantages for classification prediction of community nodes, characteristics of business objects are transmitted among the nodes through edges among the nodes, the more the nodes are associated, the richer the characteristics are transmitted, and accordingly, the community mining result is optimized.
The classification model takes the feature vector of single service data as input, does not consider the relevance among service objects, has good classification effect on isolated nodes without relevance or with sparse relevance, but has poor classification effect on service objects with rich relevance due to neglect of the relevance.
For the same category, considering the influence of the degree of association between the business objects represented by the node degree on the graph neural network and the classification model, the degree of association between the business objects represented by the node degree can be used as a parameter for adjusting the first probability and the second probability, and the first probability and the second probability are linearly or nonlinearly merged into a third probability through the node degree, wherein the third probability is a result of comprehensively measuring the classification of the graph neural network and a classification model.
In one embodiment of the present application, step 105 may include the steps of:
step 1051, respectively calculating a first weight of the neural network of the graph for the classification and a second weight of the classification model for the classification according to the node degrees.
Considering the influence of the degree of association between the node degree representation and the business objects on the graph neural network and the classification model, the weight which is suitable for classification and can be configured on the graph neural network by referring to the degree of association between the node degree representation and the business objects is recorded as a first weight, the first weight can reflect the advantages of the graph neural network on the classification of the community nodes and the disadvantages of the classification of the isolated nodes, and similarly, the weight which is suitable for classification and can be configured on the classification model by referring to the degree of association between the node degree representation and the business objects is recorded as a second weight, and the second weight can reflect the advantages of the classification model on the classification of the community nodes and the advantages of the classification of the isolated nodes.
In a specific implementation, a first mapping function may be configured in advance for the graph neural network according to a specific service requirement in a service scenario, and a second mapping function may be configured for the classification model.
Then, when the first weight is configured for the class classified by the neural network and the second weight is configured for the class classified by the classification model, the node degree may be substituted into the first mapping function configured for the graph neural network to generate the first weight for the class by the graph neural network and the node degree may be substituted into the second mapping function configured for the classification model to generate the second weight for the class by the second mapping function.
The first mapping function and the second mapping function usually belong to monotonously increasing functions, the first weight is positively correlated with the node degree, namely the node degree is larger, the first weight is larger, otherwise, the node degree is smaller, the first weight is smaller, the second weight is positively correlated with the node degree, namely, the node degree is larger, the second weight is larger, and otherwise, the node degree is smaller, and the second weight is smaller.
The first mapping function and the second mapping function are designed in pairs, the increasing rate of the first weight is not consistent with the increasing rate of the second weight, and for the same node degree, the first weight and the second weight are different, so that the importance between the neural network of the graph and the classification model is different, and the method is suitable for different service scenes.
If the node degree indicates that the node is a community node, the first weight is larger than or equal to the second weight so as to represent the importance of the graph neural network to the community node.
And if the node degree indicates that the node is an isolated node, the first weight is smaller than the second weight so as to represent the importance of the classification model to the isolated node.
In one example, the first mapping function includes:
wherein HG(x) The first weight is x, the node degree is x, δ is a lower limit value of the weight, δ is epsilon [0,1), and α is a hyperparameter, such as α ═ 1.
Accordingly, the second mapping function includes:
wherein HP(x) And x is the node degree, δ is the lower limit value of the weight, δ ∈ [0,1), and α is a hyperparameter, for example, α is 1.
In this example, the weights (first weight, second weight) are better adapted to the specific traffic needs of certain traffic scenarios with a slow smooth increase of the node degree.
In another example, the first mapping function includes:
wherein HG(x) The first weight is x, the node degree is x, δ is a lower limit value of the weight, δ ∈ [0,1), and β is a hyperparameter, for example, β ═ 1.
Accordingly, the second mapping function includes:
wherein HP(x) Is the second weight, x is the node degree, δ is the weightThe lower limit value of weight, δ ∈ [0,1), β, γ are both superparameters, such as β ═ 1, γ ═ 2.
In this example, there is a range of node degrees within which the weights (first weight, second weight) grow to quickly better fit the specific traffic needs of certain traffic scenarios.
In yet another example, the first mapping function includes:
wherein HG(x) Is the first weight, x is the node degree, δ is the lower limit value of the weight, δ ∈ [0, 1).
Accordingly, the second mapping function comprises:
wherein HP(x) For the second weight, x is the node degree, δ is the lower limit value of the weight, and μ and ∈ belong to the superparameters, such as μ ═ 2 and ∈ ═ 2.
In this example, there is a range of node degrees within which the weights (first weight, second weight) grow to quickly better fit the specific traffic needs of certain traffic scenarios.
Of course, the first mapping function and the second mapping function are only examples, and when the embodiment is implemented, other first mapping functions and other second mapping functions may be set according to actual situations, which is not limited in the embodiment of the present application. In addition, besides the first mapping function and the second mapping function, those skilled in the art may also adopt other first mapping functions and second mapping functions according to actual needs, which is not limited in this embodiment.
Step 1052, setting the product between the first weight and the first probability as a first tuning value, and setting the product between the second weight and the second probability as a second tuning value for the same category.
And 1053, calculating the sum of the first adjusting value and the second adjusting value as a third probability.
In this embodiment, the first probability and the second probability are fused into the third probability in a linear manner, that is, for the same class, for the graph neural network, the first weight and the first probability are multiplied, the obtained product is recorded as a first tuning value, for the classification model, the second weight and the second probability are multiplied, the obtained product is recorded as a second tuning value, the first tuning value and the second tuning value are added, and the obtained sum value is recorded as the third probability.
Suppose, the neural network of the graph predicts nodes viThe first probability of belonging to each class is
To node viConfiguring a first weight
Classification model prediction node viThe second probability of belonging to each class is
To node viConfiguring a second weight
Then the integration node viThe third probability of belonging to each class is
And 106, determining the class to which the business object belongs according to the third probability.
In this embodiment, a rule may be designed for classification according to the confidence level in advance, and if the third probability of a certain class satisfies the rule, which indicates that the confidence level that the business object belongs to the class is higher, it may be finally determined that the business object belongs to the class.
Illustratively, the largest one of the third probabilities of the categories is selected as a target probability, and the category corresponding to the target probability is determined as the category to which the business object belongs.
Of course, the above method for determining the category is only an example, and when the embodiment is implemented, other methods for determining the category may be set according to actual situations, for example, a maximum one is selected from third probabilities greater than a probability threshold, and is used as a target probability, and the category corresponding to the target probability is determined as a category to which a business object belongs, and the like, which is not limited in this embodiment of the application. In addition, besides the above method for determining the category, a person skilled in the art may also use other methods for determining the category according to actual needs, and the embodiment also does not limit this.
In the embodiment, the business object is converted into a graph neural network, the business object has a plurality of eigenvectors, and nodes in the graph neural network represent the plurality of eigenvectors of the business object; calculating the degree of association between nodes in the graph neural network as the node degree of the nodes; executing the graph neural network to output a first probability that the business object belongs to a preset category; executing a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector; for the same category, the first probability and the second probability are fused into a third probability according to the node degree; and determining the class to which the business object belongs according to the third probability. The graph neural network has advantages for the classification prediction of the business objects with strong correlation degree and has disadvantages for the classification prediction of the business objects with weak correlation degree, and the classification model has advantages for the classification prediction of the business objects with weak correlation degree, the classification prediction of the business objects with strong association degree has disadvantages, the two have complementary relation, the comprehensiveness of the classification effect can be ensured, the influence of the association degree between the business objects represented by the node degree on the graph neural network and the classification model is considered, the classification result of the neural network of the graph and the classification result of the classification model can be fused according to the degree of association between business objects represented by the node degree, the node degree is different under different business scenes, the classification result of the fusion graph neural network and the classification result of the classification model can be flexibly adjusted, and the classification precision is greatly improved.
Example two
Fig. 3 is a schematic structural diagram of a classification apparatus for business objects according to a second embodiment of the present application. As shown in fig. 3, the apparatus includes:
a graph neural network conversion module 301, configured to convert a business object into a graph neural network, where the business object has a plurality of feature vectors, and nodes in the graph neural network represent the plurality of feature vectors of the business object;
a node degree calculation module 302, configured to calculate a degree of association between the nodes in the graph neural network as a node degree of the nodes;
a first classification module 303, configured to execute the graph neural network to output a first probability that the business object belongs to a preset category;
a second classification module 304, configured to execute a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector;
a probability fusion module 305, configured to fuse, according to the node degree, the first probability and the second probability into a third probability for the same category;
a category determining module 306, configured to determine the category to which the service object belongs according to the third probability.
In one embodiment of the present application, the node degree calculation module 302 includes:
and the edge counting module is used for counting the number of edges connected with the nodes for each node in the graph neural network to serve as the node degree of the node, wherein one edge is connected with two nodes.
In one embodiment of the present application, the edge statistics module includes:
a node degree initialization module, configured to initialize a node degree to zero for each node in the graph neural network;
the edge query module is used for querying edges in the graph neural network;
and the node degree accumulation module is used for accumulating one to the node degrees of the two nodes connected by the edge aiming at each edge in the graph neural network.
In one embodiment of the present application, the probability fusion module 305 includes:
the weight calculation module is used for respectively calculating a first weight of the graph neural network for classification and a second weight of the classification model for classification according to the node degrees;
a tuning value calculation module configured to set, for the same category, a product between the first weight and the first probability as a first tuning value, and a product between the second weight and the second probability as a second tuning value;
and the modulation value summing module is used for calculating a sum value between the first modulation value and the second modulation value as a third probability.
In one embodiment of the present application, the weight calculation module includes:
a first mapping module for substituting the node degrees into a first mapping function configured for the graph neural network to generate a first weight of the graph neural network for a classification;
a second mapping module, configured to substitute the node degrees into a second mapping function configured for the classification model to generate a second weight of the second mapping function for the classification;
wherein the first weight is positively correlated with the node degree, and the second weight is positively correlated with the node degree;
if the node degree indicates that the node is a community node, the first weight is larger than or equal to the second weight;
if the node degree indicates that the node is an isolated node, the first weight is smaller than the second weight.
In one example of this embodiment, the first mapping function includes:
wherein HG(x) The first weight is used as the first weight, x is the node degree, delta is the lower limit value of the weight, and alpha is a hyperparameter;
the second mapping function includes:
wherein HP(x) And taking the second weight as the reference value, wherein x is the node degree, delta is the lower limit value of the weight, and alpha is a hyperparameter.
In another example of this embodiment, the first mapping function includes:
wherein HG(x) The first weight is used as the reference value, x is the node degree, delta is the lower limit value of the weight, and beta is a hyperparameter;
the second mapping function comprises:
wherein HP(x) And taking the second weight as the reference value, wherein x is the node degree, delta is the lower limit value of the weight, and beta and gamma are both hyperparameters.
In yet another example of this embodiment, the first mapping function includes:
wherein HG(x) The first weight is used as the index, x is the node degree, and delta is the lower limit value of the weight;
the second mapping function includes:
wherein HP(x) And the second weight is taken as the index, x is the node degree, delta is the lower limit value of the weight, and mu and epsilon belong to the hyperparameters.
In one embodiment of the present application, the category determination module 306 includes:
a target probability selection module for selecting the largest one from the third probabilities of the respective categories as a target probability;
and the target probability determining module is used for determining that the category corresponding to the target probability is the category to which the business object belongs.
The classification device for the business object provided by the embodiment of the application can execute the classification method for the business object provided by any embodiment of the application, and has the corresponding functional modules and beneficial effects of executing the classification method for the business object.
EXAMPLE III
Fig. 4 shows a schematic structural diagram of a classification device 10 of business objects that can be used to implement an embodiment of the present application.
As shown in fig. 4, the classification device 10 of the business object includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a Read Only Memory (ROM)12, a Random Access Memory (RAM)13, and the like, wherein the memory stores a computer program executable by the at least one processor, and the processor 11 can perform various suitable actions and processes according to the computer program stored in the Read Only Memory (ROM)12 or the computer program loaded from a storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data necessary for the operation of the classification device 10 of the business object can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to the bus 14.
A number of components in the classification device 10 of the business object are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, or the like; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, an optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the classification device 10 of the business object to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The processor 11 performs the various methods and processes described above, such as the classification method of the business object.
In some embodiments, the classification method of the business object may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed on the classification device 10 of a business object via the ROM 12 and/or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the classification method of a business object described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the classification method of the business object by any other suitable means (e.g., by means of firmware).
Claims (12)
1. A method for classifying business objects, comprising:
converting a business object into a graph neural network, the business object having a plurality of feature vectors, nodes in the graph neural network representing the plurality of feature vectors of the business object;
calculating the association degree between the nodes in the graph neural network as the node degree of the nodes;
executing the graph neural network to output a first probability that the business object belongs to a preset category;
executing a preset classification model to identify a second probability that the business object belongs to a preset class by using the feature vector;
for the same category, fusing the first probability and the second probability into a third probability according to the node degree;
and determining the class to which the business object belongs according to the third probability.
2. The method according to claim 1, wherein the calculating, in the graph neural network, the degree of association between the nodes as the node degrees of the nodes comprises:
and counting the number of edges connected with the nodes as the node degrees of the nodes aiming at each node in the graph neural network, wherein one edge is connected with two nodes.
3. The method of claim 2, wherein the counting, for each node in the graph neural network, the number of edges connected by the node as the node degree of the node comprises:
initializing a node degree of zero for each of the nodes in the graph neural network;
querying edges in the graph neural network;
for each edge in the graph neural network, accumulating one for the node degrees of the two nodes connected by the edge.
4. The method according to claim 1, wherein the merging the first probability and the second probability into a third probability according to the node degree for the same category comprises:
respectively calculating a first weight of the graph neural network for classification and a second weight of the classification model for classification according to the node degrees;
setting a product between the first weight and the first probability as a first tuning value, and a product between the second weight and the second probability as a second tuning value, for the same category;
and calculating the sum of the first modulation value and the second modulation value as a third probability.
5. The method of claim 4, wherein the calculating the first weight of the neural network for classification and the second weight of the classification model for classification according to the node degrees respectively comprises:
substituting the node degrees into a first mapping function configured for the graph neural network to generate a first weight of the graph neural network for a classification;
substituting the node degrees into a second mapping function configured for the classification model to generate a second weight of the second mapping function for the classification;
wherein the first weight is positively correlated with the node degree, and the second weight is positively correlated with the node degree;
if the node degree represents that the node is a community node, the first weight is larger than or equal to the second weight;
if the node degree indicates that the node is an isolated node, the first weight is smaller than the second weight.
6. The method of claim 5,
the first mapping function comprises:
wherein HG(x) The first weight is used as the first weight, x is the node degree, delta is the lower limit value of the weight, and alpha is a hyperparameter;
the second mapping function includes:
wherein HP(x) And taking the second weight as the reference value, wherein x is the node degree, delta is the lower limit value of the weight, and alpha is a hyperparameter.
7. The method of claim 5,
the first mapping function includes:
wherein HG(x) The first weight is used as the reference value, x is the node degree, delta is the lower limit value of the weight, and beta is a hyperparameter;
the second mapping function includes:
wherein HP(x) And taking the second weight as the reference value, wherein x is the node degree, delta is the lower limit value of the weight, and beta and gamma are both hyperparameters.
8. The method of claim 5,
the first mapping function includes:
wherein HG(x) The first weight is used as the index, x is the node degree, and delta is the lower limit value of the weight;
the second mapping function includes:
wherein HP(x) And the second weight is taken as the index, x is the node degree, delta is the lower limit value of the weight, and mu and epsilon belong to the hyperparameters.
9. The method according to any of claims 1-8, wherein said determining the class to which the business object belongs according to the third probability comprises:
selecting a maximum one from the third probabilities of the respective categories as a target probability;
and determining the class corresponding to the target probability as the class to which the business object belongs.
10. An apparatus for classifying a business object, comprising:
the graph neural network conversion module is used for converting a business object into a graph neural network, wherein the business object is provided with a plurality of eigenvectors, and nodes in the graph neural network represent the plurality of eigenvectors of the business object;
the node degree calculation module is used for calculating the association degree between the nodes in the graph neural network as the node degree of the nodes;
the first classification module is used for executing the graph neural network to output a first probability that the business object belongs to a preset class;
the second classification module is used for executing a preset classification model so as to identify a second probability that the business object belongs to a preset class by using the feature vector;
a probability fusion module, configured to fuse, according to the node degree, the first probability and the second probability into a third probability for the same category;
and the class determining module is used for determining the class to which the business object belongs according to the third probability.
11. A classification device for a business object, the classification device for a business object comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of classifying a business object of any of claims 1-9.
12. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program for causing a processor to carry out the method of classifying a business object of any one of claims 1-9 when executed.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210203772.XA CN114564516A (en) | 2022-03-03 | 2022-03-03 | Business object classification method, device, equipment and storage medium |
| PCT/CN2023/077115 WO2023165352A1 (en) | 2022-03-03 | 2023-02-20 | Service object classification method and apparatus, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210203772.XA CN114564516A (en) | 2022-03-03 | 2022-03-03 | Business object classification method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114564516A true CN114564516A (en) | 2022-05-31 |
Family
ID=81718672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210203772.XA Pending CN114564516A (en) | 2022-03-03 | 2022-03-03 | Business object classification method, device, equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN114564516A (en) |
| WO (1) | WO2023165352A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023165352A1 (en) * | 2022-03-03 | 2023-09-07 | 百果园技术(新加坡)有限公司 | Service object classification method and apparatus, device and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009093B (en) * | 2018-12-07 | 2020-08-07 | 阿里巴巴集团控股有限公司 | Neural network system and method for analyzing relational network graph |
| US11551039B2 (en) * | 2020-04-28 | 2023-01-10 | Microsoft Technology Licensing, Llc | Neural network categorization accuracy with categorical graph neural networks |
| CN111667067A (en) * | 2020-05-28 | 2020-09-15 | 平安医疗健康管理股份有限公司 | Recommendation method and device based on graph neural network and computer equipment |
| CN113988264A (en) * | 2021-10-29 | 2022-01-28 | 支付宝(杭州)信息技术有限公司 | Method and device for obtaining graph neural network for executing flow prediction service |
| CN114564516A (en) * | 2022-03-03 | 2022-05-31 | 百果园技术(新加坡)有限公司 | Business object classification method, device, equipment and storage medium |
-
2022
- 2022-03-03 CN CN202210203772.XA patent/CN114564516A/en active Pending
-
2023
- 2023-02-20 WO PCT/CN2023/077115 patent/WO2023165352A1/en unknown
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023165352A1 (en) * | 2022-03-03 | 2023-09-07 | 百果园技术(新加坡)有限公司 | Service object classification method and apparatus, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023165352A1 (en) | 2023-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919316B (en) | Method, device and equipment for acquiring network representation learning vector and storage medium | |
| CN111565316B (en) | Video processing method, video processing device, computer equipment and storage medium | |
| CN110909182A (en) | Multimedia resource searching method and device, computer equipment and storage medium | |
| US20110295787A1 (en) | Information processing apparatus, information processing method, and program | |
| JP2022020070A (en) | Information processing, information recommendation method and apparatus, electronic device and storage media | |
| CN111062431A (en) | Image clustering method, image clustering device, electronic device, and storage medium | |
| CN111310025B (en) | Model training method, data processing device and related equipment | |
| Wang et al. | Latency-aware adaptive video summarization for mobile edge clouds | |
| Wang et al. | QoS prediction of web services based on reputation-aware network embedding | |
| WO2023165352A1 (en) | Service object classification method and apparatus, device and storage medium | |
| Gao et al. | A deep learning framework with spatial-temporal attention mechanism for cellular traffic prediction | |
| CN113590898A (en) | Data retrieval method and device, electronic equipment, storage medium and computer product | |
| CN114332550A (en) | Model training method, system, storage medium and terminal equipment | |
| CN112989182A (en) | Information processing method, information processing apparatus, information processing device, and storage medium | |
| Nikoloska et al. | Data selection scheme for energy efficient supervised learning at IoT nodes | |
| CN112541556A (en) | Model construction optimization method, device, medium, and computer program product | |
| Chen et al. | A user dependent web service QoS collaborative prediction approach using neighborhood regularized matrix factorization | |
| CN113705683B (en) | Training method and device of recommendation model, electronic equipment and storage medium | |
| CN111935259B (en) | Method and device for determining target account set, storage medium and electronic equipment | |
| CN112860759B (en) | Big data mining method based on block chain security authentication and cloud authentication service system | |
| CN115858911A (en) | Information recommendation method and device, electronic equipment and computer-readable storage medium | |
| CN113569067A (en) | Label classification method and device, electronic equipment and computer readable storage medium | |
| CN111666452A (en) | Method and device for clustering videos | |
| CN111881352A (en) | Content pushing method and device, computer equipment and storage medium | |
| CN111581420A (en) | Medical image real-time retrieval method based on Flink |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |