CN117670572A - Social behavior prediction method, system and product based on graph comparison learning - Google Patents
Social behavior prediction method, system and product based on graph comparison learning Download PDFInfo
- Publication number
- CN117670572A CN117670572A CN202410145286.6A CN202410145286A CN117670572A CN 117670572 A CN117670572 A CN 117670572A CN 202410145286 A CN202410145286 A CN 202410145286A CN 117670572 A CN117670572 A CN 117670572A
- Authority
- CN
- China
- Prior art keywords
- network
- social
- node
- influence
- graph
- 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.)
- Granted
Links
- 230000011273 social behavior Effects 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 230000007246 mechanism Effects 0.000 claims abstract description 24
- 230000006399 behavior Effects 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 15
- 230000004931 aggregating effect Effects 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 42
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000010586 diagram Methods 0.000 claims description 20
- 230000004913 activation Effects 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 14
- 238000003860 storage Methods 0.000 claims description 13
- 238000009826 distribution Methods 0.000 claims description 11
- 230000002776 aggregation Effects 0.000 claims description 7
- 238000004220 aggregation Methods 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 description 16
- 238000004891 communication Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 3
- 238000003012 network analysis Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000013604 expression vector Substances 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000011022 operating instruction Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a social behavior prediction method, a social behavior prediction system and a social behavior prediction product based on graph comparison learning, and relates to the field of social behavior prediction. The method comprises the following steps: determining network structure characteristics, node influence characteristics and network inherent characteristics according to the social influence network; the social influence network is established through social network data; the network structure characteristics are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent characteristics of the network are obtained by connecting the characteristics of the user configuration file, the characteristics of homogeneity in the social influence network and the characteristics of nodes when social behavior is completed in series; splicing network structure characteristics, node influence characteristics and network inherent characteristics to generate input layer characteristics; initializing the input layer characteristics into a graph neural network, and predicting social behavior of a user. The method and the device understand the behavior rules and effectively improve the social behavior prediction accuracy.
Description
Technical Field
The invention relates to the field of social behavior prediction, in particular to a social behavior prediction method, a social behavior prediction system and a social behavior prediction product based on graph comparison learning.
Background
With the popularity of social networks and the rapid growth of data, predicting social behavior is becoming increasingly important. In social networks, such as WeChat and microblog, users may interact with friends by replying to their comments or praying to the same product or service. The more friends that participate in the interaction, the more interested the user is in the product or service. More and more businesses begin to focus on participating marketing approaches that maximize social behavior.
In fact, users are connected to only a limited number of social users, so the "user-to-user" network is relatively sparse. Some research efforts are currently attempting to alleviate the sparsity of data by utilizing auxiliary information, but the effect is very limited.
Although some research efforts on social behavior prediction have alleviated the cold start problem, they have focused solely on network structural features and have not fundamentally addressed this problem. Moreover, existing research does not take into account both factors of node impact characteristics and negative sampling of the network, such as recommendation systems on social platform data. Individuals must be affected by and to neighboring nodes in a social network. It can be seen that the prediction accuracy of the existing social behavior prediction mode is low.
Disclosure of Invention
The invention aims to provide a social behavior prediction method, a social behavior prediction system and a social behavior prediction product based on graph comparison learning, so as to solve the problem of low social behavior prediction precision.
In order to achieve the above object, the present invention provides the following solutions:
a social behavior prediction method based on graph comparison learning comprises the following steps:
determining network structure characteristics, node influence characteristics and network inherent characteristics according to the social influence network; the social influence network is established through social network data; the network structural features are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent network characteristics are obtained by connecting user profile characteristics, homogeneity characteristics in the social influence network and node characteristics when social behavior is completed in series;
splicing the network structure characteristics, the node influence characteristics and the network inherent characteristics to generate input layer characteristics;
initializing the input layer characteristics into a graph neural network, and predicting social behaviors of users.
Optionally, determining the network structure feature, the node influence feature and the network inherent feature according to the social influence network specifically includes:
determining a positive sample subgraph and an anchor point sample subgraph according to the social influence network, calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph according to a graph neural network, and determining network structure characteristics;
analyzing neighbor nodes of each node on the social influence network by adopting an attention mechanism in the self-attention network, calculating hidden representation of each node, and determining node influence characteristics;
based on the social influence network, the user profile features, the homogeneity features in the social influence network and the node features when social behavior is completed are connected in series to generate network inherent features.
Optionally, determining a positive sample sub-graph and an anchor sample sub-graph according to the social influence network, and calculating neighbor characteristics of each node in the positive sample sub-graph and the anchor sample sub-graph according to a graph neural network, so as to determine network structure characteristics, which specifically includes:
randomly deleting edges in the social influence network with random probability, and generating a first network diagram corresponding to the social influence network;
Scanning the adjacent matrix of the first network graph according to the row, and calculating Bernoulli distribution of the rest edges;
determining a positive sample subgraph from the adjacency matrix and the bernoulli distribution;
randomly shielding the dimension of the node characteristic in the social influence network, and generating a second network diagram corresponding to the social influence network;
randomly scanning node characteristics of the second network graph, and generating an anchor point sample subgraph according to the random vector of 0-1 and the randomly scanned node characteristics;
calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph by using a graph neural network; the same nodes in the positive sample subgraph and the anchor point sample subgraph are used as positive examples, and different nodes are used as negative examples;
and continuously updating the characteristic representations of all nodes while continuously converging by adopting a gradient descent method, and determining the network structure characteristics.
Optionally, on the social influence network, analyzing neighbor nodes of each node by adopting an attention mechanism in a self-attention network, calculating hidden representation of each node, and determining node influence characteristics, including:
adopting an attention mechanism in a self-attention network, and constructing an attention function according to the current node characteristics and the neighbor node characteristics of the current node on the social influence network; the attention mechanism is a graph attention mechanism;
According to the attention function, carrying out normalization processing on the attention coefficient by utilizing a softmax function to generate the attention coefficient;
determining a linear combination of a nonlinear ReLU activation function and an adjacent attention network layer according to the attention coefficient;
and determining the node influence characteristic according to the linear combination.
Optionally, based on the social influence network, the user profile feature, the homogeneity feature in the social influence network and the node feature when the social behavior is completed are connected in series, and the generating of the network inherent feature specifically includes:
taking the node with the configuration file on the social influence network as a user configuration file characteristic;
capturing homogeneity characteristics in the social influence network by adopting an algorithm for efficiently learning continuous characteristics of nodes in the network;
taking the node characteristics when the social behavior is completed on the advertisement or the business as the node characteristics when the social behavior is completed;
and generating inherent characteristics of the network by using the characteristics of the user profile, the homogeneity characteristics in the social influence network and the node characteristics when the social behavior is completed by using the series symbols.
Optionally, the network structure feature, the node influence feature and the network inherent feature are spliced to generate an input layer feature, which specifically includes:
Respectively generating a feature matrix and an adjacent matrix of the positive sample subgraph and the anchor sample subgraph;
maximizing the consistency of the positive sample subgraph and the anchor point sample subgraph, deleting inconsistent nodes or edges in the two subgraphs, and representing the reserved nodes by a feature matrix; the two subgraphs are the positive sample subgraph and the anchor sample subgraph;
based on the feature matrix representation, the reserved nodes are projected to the two-layer perceptron to generate input layer features.
Optionally, initializing the input layer feature into a graph neural network, and predicting social behavior of the user, including:
processing the input layer characteristics through a multi-layer network structure in the graph neural network to perform forward propagation;
for the characteristics of each layer of aggregation domain nodes in the multi-layer network structure, generating new characteristics of the nodes, and applying a ReLU activation function to the new characteristics;
calculating a plurality of category probability values by using a softmax activation function at the last layer of the multi-layer network structure, and determining a final predicted value;
and updating the gradient into the weight of the graph neural network through back propagation, and selecting the category corresponding to the final predicted value as the predicted social behavior of the user.
A graph-contrast learning-based social behavior prediction system, comprising:
the feature determining module is used for determining network structure features, node influence features and network inherent features according to the social influence network; the social influence network is established through social network data; the network structural features are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent network characteristics are obtained by connecting user profile characteristics, homogeneity characteristics in the social influence network and node characteristics when social behavior is completed in series;
the input layer characteristic generation module is used for splicing the network structure characteristic, the node influence characteristic and the network inherent characteristic to generate an input layer characteristic;
and the social behavior prediction module is used for initializing the input layer characteristics into the graphic neural network and predicting the social behavior of the user.
An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the social behavior prediction method based on graph contrast learning described above.
Optionally, the memory is a non-transitory computer readable storage medium, and the non-transitory computer readable storage medium stores a computer program, and when the computer program is executed by the processor, the social behavior prediction method based on graph comparison learning is implemented.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the method comprises the steps of determining network structure characteristics, node influence characteristics and network inherent characteristics based on a social influence network; the network structure characteristics are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent network characteristics are obtained by connecting user profile characteristics, homogeneity characteristics in the social influence network and node characteristics when social behavior is completed in series; and combining the network structure characteristics, the node influence characteristics and the network inherent characteristics as input layer characteristics of the graph neural network so as to predict social behaviors of the user. According to the method, the network structure characteristics are fully mined, the node influence characteristics obtained by combining the characteristics of the neighbor nodes and the behavior state learning are introduced, the influence of the characteristics of the neighbor nodes is considered, and the prediction accuracy of social behaviors is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a social behavior prediction method based on graph comparison learning provided by the invention;
FIG. 2 is a flow chart of the structure characteristics of the graph contrast learning pre-training network provided by the invention;
FIG. 3 is a diagram of an example neighborhood of partial nodes of an Epinions social network extracted after graph comparison learning.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a social behavior prediction method, a social behavior prediction system and a social behavior prediction product based on graph comparison learning, which can improve the prediction accuracy of social behaviors of users.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Example 1
The invention provides a social behavior prediction method based on graph comparison learning, which comprises the following steps:
determining network structure characteristics, node influence characteristics and network inherent characteristics according to the social influence network; the node influence characteristics are learned by combining the characteristics and behavior states of the neighbor nodes.
And splicing the network structure characteristic, the node influence characteristic and the network inherent characteristic to generate an input layer characteristic.
Initializing the input layer characteristics into a graph neural network, and predicting social behaviors of users.
In practical application, determining network structure characteristics, node influence characteristics and network inherent characteristics according to a social influence network specifically comprises: determining a positive sample subgraph and an anchor point sample subgraph according to the social influence network, calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph according to a graph neural network, and determining network structure characteristics; analyzing neighbor nodes of each node on the social influence network by adopting an attention mechanism in the self-attention network, calculating hidden representation of each node, and determining node influence characteristics; based on the social influence network, the user profile features, the homogeneity features in the social influence network and the node features when social behavior is completed are connected in series to generate network inherent features.
In practical application, as shown in fig. 1, the prediction method provided by the present invention is divided into five steps.
Step one: the network structural features are pre-trained.
The network structure characteristics are determined by the following steps: randomly deleting edges in the social influence network with random probability, and generating a first network diagram corresponding to the social influence network; scanning the adjacent matrix of the first network graph according to the row, and calculating Bernoulli distribution of the rest edges; determining a positive sample subgraph from the adjacency matrix and the bernoulli distribution; randomly shielding the dimension of the node characteristic in the social influence network, and generating a second network diagram corresponding to the social influence network; randomly scanning node characteristics of the second network graph, and generating an anchor point sample subgraph according to the random vector of 0-1 and the randomly scanned node characteristics; calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph by using a graph neural network; the same nodes in the positive sample subgraph and the anchor point sample subgraph are used as positive examples, and different nodes are used as negative examples; and continuously updating the characteristic representations of all nodes while continuously converging by adopting a gradient descent method, and determining the network structure characteristics.
The determination of network structural characteristics is applied to a specific embodiment.
(1) At a given social influence network graphOn, with random probability->Deleting some edges in the original network to obtain a first network diagram +.>。
(2) Scanning by lineFor existing edges, their bernoulli distribution is calculated as follows:
wherein->The expression is represented by->Generated Bernoulli distribution, ->Representing a bernoulli distribution generating function.
(3) By usingAdjacent matrix and->Calculating Hadamard product (+)>) I.e. generate->Is->Considered as positive sample, +.>Positive sample subgraphs for randomly deleted edges; />Is an adjacency matrix; />Is the Hadamard product.
(4) At a given social influence network graphOn top of that, some dimensions of some node features are randomly masked with 0, resulting in a second network graph +.>。
(5) Random scanningIs calculated by using the random 0-1 vector and the node characteristics, namely, a second network diagram is generated +.>Is->Considered as anchor sample, ++>Subgraphs that are randomly hidden node features.
(6) Use map auto encoder pairAnd->Coding is performed separately to obtain node expression vectors of the nodes, and a contrast loss function (gradient descent method) is used to calculate the contrast loss of the expression vectors. When the loss value is minimum, the positive samples are close to each other, the negative samples are far away from each other, and the graph structure of the network, namely the output network structure characteristics, are determined by the positive and negative samples. The contrast loss function used is as follows:
。
Wherein,the expression parameter is->Activation probability of->Representation->Includes node->Is->Indicate->Personal node->Is in->Time of day activation status->For the i node->Is in->Time of day activation status->Is a time increment.
Step two: the pre-training node affects the force characteristics.
The node influence characteristic determination process comprises the following steps: adopting an attention mechanism in a self-attention network, and constructing an attention function according to the current node characteristics and the neighbor node characteristics of the current node on the social influence network; the attention mechanism is a graph attention mechanism; according to the attention function, carrying out normalization processing on the attention coefficient by utilizing a softmax function to generate the attention coefficient; determining a linear combination of a nonlinear ReLU activation function and an adjacent attention network layer according to the attention coefficient; and determining the node influence characteristic according to the linear combination.
The process of determining the node influence characteristics is applied in a specific embodiment.
(1) On a given social impact network, in order to better integrate the correlation between node features into the model, attention mechanisms in the graph attention network (GAT) are employed to aggregate the impact of their first and second order neighbor nodes to obtain a more robust feature representation. With the graph attention mechanism, the attention function is as follows:
。
Wherein e ij In order for the attention coefficient to be a factor of attention,is a weight vector, +.>Weight matrix (W/W)>And->Respectively is node->And neighbor node->Is characterized by (1)>Representing vector transpose->Representing the vector concatenation, the LeakyReLU is the activation function.
(2) Normalization of the attention coefficient with softmax yields the attention coefficient. Wherein (1)>Representing node->Neighbor set of->Expressed in the collection->Is included in the node (a).
(3) The influence pre-training model of the self-attention network is realized by using the linear combination of the nonlinear ReLU activation function and the adjacent GAT layer, and the linear combination formula is as follows:the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Indicate->Layer (S)>Representing node->In network->Layer characteristics,/->Representing a combined function->The j-th node represented in a given social influence network graph is included in the set +.>Is included in the node (a).
(4) In GAT, the last layer (the firstLayer) is the node influence feature representation +.>。
Step three: extracting network inherent characteristics.
The determination process of the inherent characteristics of the network comprises the following steps: taking the node with the configuration file on the social influence network as a user configuration file characteristic; capturing homogeneity characteristics in the social influence network by adopting an algorithm for efficiently learning continuous characteristics of nodes in the network; taking the node characteristics when the social behavior is completed on the advertisement or the business as the node characteristics when the social behavior is completed; and generating inherent characteristics of the network by using the characteristics of the user profile, the homogeneity characteristics in the social influence network and the node characteristics when the social behavior is completed by using the series symbols.
The determination of the network inherent characteristics is applied in a specific embodiment.
(1) On a given social influence network, the user profile features are node-specific features of the first class. For nodes with similar profiles, the attention factor will be higher.
(2) Inputting a given social influence network into a node2vec algorithm (an algorithm for efficiently learning the continuous features of nodes in the network) results in a low-dimensional feature representation of the graph nodes, i.e. a second class of features.
(3) The third type of feature is a node feature when social behavior is completed on recent advertisements or services, that is, adding a dimension to the behavior on one advertisement or service.
(4) The features obtained in the steps (1), (2) and (3) are marked by symbolsTandem into one feature, which is considered as a network inherent feature, namely:
first kind of characteristicsSecond kind of features->A third class of features.
Step four: and (3) fusing the characteristic representations obtained in the step one to the step three to obtain the input layer characteristics.
The splicing process is as follows: respectively generating a feature matrix and an adjacent matrix of the positive sample subgraph and the anchor sample subgraph; maximizing the consistency of the positive sample subgraph and the anchor point sample subgraph, deleting inconsistent nodes or edges in the two subgraphs, and representing the reserved nodes by a feature matrix; the two subgraphs are the positive sample subgraph and the anchor sample subgraph; based on the feature matrix representation, the reserved nodes are projected to the two-layer perceptron to generate input layer features.
The above described stitching process is applied to specific embodiments.
(1) In each iteration, two subgraphs (i.e., positive sample subgraphs with randomly deleted edges) are created using the method in step oneAnchor sample subgraph of random hidden node feature>) And generating a feature matrix and an adjacency matrix for each sub-graph.
(2) Maximizing the consistency of the two subgraphs, deleting inconsistent nodes or edges (i.e. unimportant nodes or edges) in the two subgraphs, and keeping important nodes and edges and representing the kept nodes by vectors.
(3) Respectively projecting the feature matrix and the adjacent matrix of the reserved nodes to a two-layer sensorLayer(s)Layers), each pair->Is defined as:
。
wherein,representing node->Respectively at->Layer and->Loss of layer after projection, +.>Representing criticizing parameters>Representing the number of nodes->Representing a temporary parameter. e is a natural constant, u i Is->Layer i node, u k Is->Layer k node, v k Is->The k-th node of the layer, i and k (node numbers) are smaller than or equal to the number N of the reserved nodes.
(4) Inputting the pre-trained network structure characteristics, the node influence characteristics and the network inherent characteristics in the previous three steps into the sensor in the step (3), and outputting the characteristics after the three characteristics are fused, wherein the characteristics have the capability of enhancing the characteristic expression.
Step five: and training a prediction model.
The prediction process of the social behavior of the user is as follows: processing the input layer characteristics through a multi-layer network structure in the graph neural network to perform forward propagation; for the characteristics of each layer of aggregation domain nodes in the multi-layer network structure, generating new characteristics of the nodes, and applying a ReLU activation function to the new characteristics; calculating a plurality of category probability values by using a softmax activation function at the last layer of the multi-layer network structure, and determining a final predicted value; and updating the gradient into the weight of the graph neural network through back propagation, and selecting the category corresponding to the final predicted value as the predicted social behavior of the user.
The prediction process of the social behavior of the user is applied to the specific embodiment.
Initializing a graph neural network based on the input layer characteristics obtained in the step four; processing the characteristics of an input layer through a multi-layer network structure, and carrying out forward propagation; aggregating the characteristics of the nodes in the field to obtain new characteristics of the nodes, and adopting the following loss function, wherein the formula is as follows:
。
wherein,tag actual value representing the i-th sample, < +.>Tag prediction value representing the i-th sample, < +.>The number of representative nodes is also the number of samples. And (5) reversely solving the parameters of each neuron by adopting gradient descent, and updating the parameters to the neural network. Repeating the steps to find the minimum value of the loss function loss. At this time, the obtained probability value is the behavior prediction result of the user.
According to the method, social network data (the data set used by the method is Epinions, flickr and newwave microblogs) can be fully mined, user behaviors and interactions in the social network can be better analyzed and understood, structural information and non-structural information in the social network can be effectively captured, interests and behaviors of users can be accurately predicted, social influence and negative sampling problems which are not considered by other methods are solved, accuracy and reliability of social behavior prediction are improved, social network analysis and recommendation systems are promoted, personalized recommendation is realized, and social network services are improved.
Example two
Based on the social behavior prediction method based on graph comparison learning provided by the embodiment one, the invention also provides a specific algorithm of the social behavior prediction method based on graph comparison learning, as follows.
Predicting social behavior by adopting graph contrast learning; firstly, two methods for extracting subgraphs from an original graph are designed, and network structural features of the subgraphs are learned through graph comparison learning; modeling how the behavior of the user is influenced by the neighbors, and learning the influence characteristics of the nodes through a graph attention network; finally, social behavior is predicted in combination with network structural features, node influence features and network inherent features.
1. The network architecture features are pre-trained as shown in fig. 2.
The pre-training network structure features are the core steps of the present invention. With random probabilityDeleting social influence network graphsObtaining a first network graph +.>. Scanning by row->For which the bernoulli distribution is calculated for the existing edges. Use->Adjacent matrix and->(by->The generated bernoulli distribution) to calculate the hadamard product, i.e. to generate the first network diagram +.>Is->(the specific process is described by algorithm 1). Random mask using 0->Some dimensions of some node features in the network, get a second network diagram +.>. Randomly scanning certain node features with random 0-1 vector and +.>The eigenvector calculates the Hadamard product, i.e. a second network diagram is generated +.>Is->(the specific process is described by algorithm 2). Coding using a graph auto-encoder to obtain +.>And->The nodes of (a) represent vectors, and the contrast loss of the represented vectors is calculated by using a contrast loss function (gradient descent method). When the loss value is minimum, the positive samples are close to each other, the negative samples are far away from each other, and the graph structure of the network, namely the output network structure characteristics, are determined by the positive and negative samples. The partial subgraph corresponding to the network structure characteristics after the epinits data set is pre-trained is shown in fig. 3, wherein black solid dots represent target users on the epinits data set, gray solid dots represent other users except the target users in the subgraph, positive influence structures are above the dotted line, and negative influence structures are below the dotted line.
Algorithm 1: the edges are randomly deleted to generate a subgraph.
Input: social influence network diagramAdjacency matrix->Random probability->
And (3) outputting:adjacency matrix of->
01: for do
02: if then
03:
04: else
05: end if
06: end for
07: return
Wherein E representsEdge set of->Representing Bernoulli distribution function, < >>Representing the bernoulli matrix.
Algorithm 2: the node features are randomly hidden to generate a subgraph.
Input: social influence network diagramFeature matrix->Random probability->
And (3) outputting:feature matrix +.>
01: initialization of
02：
03：for do
04:
05: end for
Wherein,represents a F-dimensional 0-1 random vector, F represents +.>The number of elements in->Representation vector->I element of (a)>Representing a feature matrix->The i-th element of a vector in the middle row,/->Is a feature matrix->Middle row a certain vector.
2. The pre-training node affects the force characteristics.
The method comprises the steps that a node is influenced by the behavior of a neighbor node of the node, the neighbor node is influenced at the same time, the step models how the behavior of a user is influenced by the neighbor, and the influence characteristics of the node are learned through a graph attention network. The specific process is as follows.
(1) In order to better integrate the relevance between node features into a model on a given social impact network, attention mechanisms in a graph attention network (GAT) are employed to aggregate the information of its first and second order neighbor nodes to obtain a more robust feature representation. The aggregation once can be regarded as considering the influence of the first-order neighbor nodes, and the aggregation twice simultaneously considers the influence of the first-order neighbor nodes and the second-order neighbor nodes. However, the number of aggregation times cannot be exceeded, and multiple aggregation can lead to the introduction of excessive noise in the characterization of the current node, which can have a negative impact. With the graph attention mechanism, the attention function is as follows:
。
(2) Normalization of the attention coefficient with softmax yields the attention coefficient。
(3) The influence pre-training model of the self-attention network is realized by linear combination of a nonlinear ReLU activation function and an adjacent GAT layer.
(4) In GAT, the output of the last layer is a node influence feature representation.
3. Extracting network inherent characteristics.
Network-inherent characteristics are formed at the time of network generation. The specific process of extracting the inherent characteristics of the network is as follows.
(1) The user profile feature is a node-specific feature of the first class. For nodes with similar profiles, the attention factor will be higher.
(2) Inputting a given social influence network into a node2vec algorithm (an algorithm for efficiently learning the continuous features of nodes in the network) results in a low-dimensional feature representation of the graph nodes, i.e. a second class of features.
(3) The third type of feature is a node feature when social behavior is completed on recent advertisements or services, that is, adding a dimension to the behavior on one advertisement or service.
(4) The features obtained in the steps (1), (2) and (3) are marked with symbolsTandem into one feature, which is considered as a network inherent feature, namely:
first kind of characteristics Second kind of features->A third class of features.
4. And fusing the characteristic representations obtained in the three steps to obtain the input layer characteristics.
Pre-training input layer features is a key step of the present invention. In each iteration, two subgraphs are created with algorithm 1 and algorithm 2, and then the feature matrix and adjacency matrix for each subgraph are generated. Maximizing the consistency of the two subgraphs, deleting inconsistent nodes or edges (i.e. unimportant nodes or edges) in the two subgraphs, keeping important nodes and edges, and representing the kept nodes by a feature matrix. And fusing the pre-trained network structure characteristics, the node influence characteristics and the network inherent characteristics to form the input layer characteristics of the prediction model. See algorithm 3 for specific procedures.
Algorithm 3: the input layer features are pre-trained.
Inputting network structure characteristics, node influence characteristics and network inherent characteristics
Output of embedded features
01 initializing ///>Representing criticizing parameters
02: calculation ofAnd-> ///>And->Respectively represent node->The loss after projection is +.>Vertex set and edge set of subgraphs of (a)
03：for epoch=1N times of/iteration
04: will randomlyIs divided into ∈10>Is->Parts, part i is->
05： for each do
06 obtaining a first network subgraph using Algorithm 1 And a second network diagram->
07： ///>Representing a set of negative sampling samples
08： ///>Representing a set of positively sampled samples
09: temporarily weight theAdded to two subgraphs
10: calculating contrast loss of two subgraphs
11: calculation of gradient intensity for each edge using back propagation of losses
12: further transforming two sub-graphs with a contrast transformation
13: removing weights from two subgraphs
14: again calculate the contrast loss of the two subgraphs
15: updating by random gradient descentValue of
16： end for
17：end for
18: splice node influence features
19: inherent features of spliced networks
5. And training a prediction model.
(1) The weights and biases of the graph neural network are initialized. Inputting the trained input layer characteristics into the graph neural network, aggregating the characteristics of the domain nodes for each layer in the network to obtain new characteristics of the nodes, and then applying the ReLU activation function to the new characteristics. At the last layer, the probability values of the multiple categories are calculated by using the defined softmax activation function, and then the final predicted value is obtained.
(2) The following loss function is calculated and the predicted tag value shows how far from the actual target the predictions of the present invention are, avoiding overfitting as much as possible.
。
Wherein, Tag actual value representing the i-th sample, < +.>Tag prediction value representing the i-th sample, < +.>The number of representative nodes is also the number of samples.
(3) And (5) reversely solving the parameters of each neuron by adopting gradient descent, and updating the parameters to the neural network.
(4) Repeating (1), (2) and (3) to find the minimum value of the loss function loss. At this time, the probability value obtained in (1) is the behavior prediction result of the user.
The invention effectively solves a series of problems that other methods do not fully excavate network structural features, do not utilize social influence, do not consider negative sampling and the like. The method can improve the accuracy and reliability of social behavior prediction, and provides more accurate data support for social network analysis and recommendation systems; the method can improve personalized recommendation effect and user experience, and provide more personalized and accurate goods and services for users; the method can improve the quality and efficiency of the social network service and provide better social network experience for the user. Meanwhile, the method and the system can bring actual benefits and progress to the fields of social network analysis, personalized recommendation, electronic commerce and the like.
Example III
In order to execute the corresponding method of the above embodiment to achieve the corresponding functions and technical effects, a social behavior prediction system based on graph comparison learning is provided below.
A graph-contrast learning-based social behavior prediction system, comprising:
the feature determining module is used for determining network structure features, node influence features and network inherent features according to the social influence network; the social influence network is established through social network data; the node influence characteristics are learned by combining the characteristics and behavior states of the neighbor nodes.
And the input layer characteristic generation module is used for splicing the network structure characteristic, the node influence characteristic and the network inherent characteristic to generate an input layer characteristic.
And the social behavior prediction module is used for initializing the input layer characteristics into the graphic neural network and predicting the social behavior of the user.
Example IV
The embodiment of the invention provides an electronic device which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor runs the computer program to enable the electronic device to execute the social behavior prediction method based on graph comparison learning.
In practical applications, the electronic device may be a server.
In practical applications, the electronic device includes: at least one processor (processor), memory (memory), bus, and communication interface (Communications Interface).
Wherein: the processor, communication interface, and memory communicate with each other via a communication bus.
And the communication interface is used for communicating with other devices.
And a processor, configured to execute a program, and specifically may execute the method described in the foregoing embodiment.
In particular, the program may include program code including computer-operating instructions.
The processor may be a central processing unit, CPU, or specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the electronic device may be the same type of processor, such as one or more CPUs; but may also be different types of processors such as one or more CPUs and one or more ASICs.
And the memory is used for storing programs. The memory may comprise high-speed RAM memory or may further comprise non-volatile memory, such as at least one disk memory.
Based on the description of the embodiments above, embodiments of the present application provide a storage medium having stored thereon computer program instructions executable by a processor to implement the method of any of the embodiments.
The social behavior prediction system based on graph comparison learning provided by the embodiment of the application exists in various forms, including but not limited to:
(1) A mobile communication device: such devices are characterized by mobile communication capabilities and are primarily aimed at providing voice, data communications. Such terminals include: smart phones, multimedia phones, and functional phones, etc.
(2) Ultra mobile personal computer device: such devices are in the category of personal computers, having computing and processing functions, and generally having mobile internet access capabilities. Such terminals include: PDA, MID, and UMPC devices, etc.
(3) Portable entertainment device: such devices may display and play multimedia content. The device comprises: audio, video players, palm game players, electronic books, and smart toys and portable car navigation devices.
(4) Other electronic devices with data interaction functions.
Thus, particular embodiments of the present subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may be advantageous.
The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functions of each element may be implemented in one or more software and/or hardware elements when implemented in the present application. It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by the computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular transactions or implement particular abstract data types. The application may also be practiced in distributed computing environments where transactions are performed by remote processing devices that are connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (10)
1. The social behavior prediction method based on graph comparison learning is characterized by comprising the following steps of:
determining network structure characteristics, node influence characteristics and network inherent characteristics according to the social influence network; the social influence network is established through social network data; the network structural features are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent network characteristics are obtained by connecting user profile characteristics, homogeneity characteristics in the social influence network and node characteristics when social behavior is completed in series;
Splicing the network structure characteristics, the node influence characteristics and the network inherent characteristics to generate input layer characteristics;
initializing the input layer characteristics into a graph neural network, and predicting social behaviors of users.
2. The social behavior prediction method based on graph comparison learning according to claim 1, wherein the determining network structure characteristics, node influence characteristics and network inherent characteristics according to the social influence network specifically comprises:
determining a positive sample subgraph and an anchor point sample subgraph according to the social influence network, calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph according to a graph neural network, and determining network structure characteristics;
analyzing neighbor nodes of each node on the social influence network by adopting an attention mechanism in the self-attention network, calculating hidden representation of each node, and determining node influence characteristics;
based on the social influence network, the user profile features, the homogeneity features in the social influence network and the node features when social behavior is completed are connected in series to generate network inherent features.
3. The social behavior prediction method based on graph comparison learning according to claim 2, wherein determining a positive sample sub-graph and an anchor sample sub-graph according to the social influence network, and calculating neighbor features of each node in the positive sample sub-graph and the anchor sample sub-graph according to a graph neural network, and determining network structure features specifically comprises:
Randomly deleting edges in the social influence network with random probability, and generating a first network diagram corresponding to the social influence network;
scanning the adjacent matrix of the first network graph according to the row, and calculating Bernoulli distribution of the rest edges;
determining a positive sample subgraph from the adjacency matrix and the bernoulli distribution;
randomly shielding the dimension of the node characteristic in the social influence network, and generating a second network diagram corresponding to the social influence network;
randomly scanning node characteristics of the second network graph, and generating an anchor point sample subgraph according to the random vector of 0-1 and the randomly scanned node characteristics;
calculating neighbor characteristics of each node in the positive sample subgraph and the anchor point sample subgraph by using a graph neural network; the same nodes in the positive sample subgraph and the anchor point sample subgraph are used as positive examples, and different nodes are used as negative examples;
and continuously updating the characteristic representations of all nodes while continuously converging by adopting a gradient descent method, and determining the network structure characteristics.
4. The social behavior prediction method based on graph comparison learning according to claim 2, wherein on the social influence network, neighbor nodes of each node are analyzed by adopting an attention mechanism in a self-attention network, hidden representation of each node is calculated, and node influence characteristics are determined, and the method specifically comprises the following steps:
Adopting an attention mechanism in a self-attention network, and constructing an attention function according to the current node characteristics and the neighbor node characteristics of the current node on the social influence network; the attention mechanism is a graph attention mechanism;
according to the attention function, carrying out normalization processing on the attention coefficient by utilizing a softmax function to generate the attention coefficient;
determining a linear combination of a nonlinear ReLU activation function and an adjacent attention network layer according to the attention coefficient;
and determining the node influence characteristic according to the linear combination.
5. The social behavior prediction method based on graph comparison learning according to claim 2, wherein generating network inherent features based on the social influence network, the series user profile features, the homogeneity features in the social influence network and the node features when social behavior is completed specifically comprises:
taking the node with the configuration file on the social influence network as a user configuration file characteristic;
capturing homogeneity characteristics in the social influence network by adopting an algorithm for efficiently learning continuous characteristics of nodes in the network;
taking the node characteristics when the social behavior is completed on the advertisement or the business as the node characteristics when the social behavior is completed;
And generating inherent characteristics of the network by using the characteristics of the user profile, the homogeneity characteristics in the social influence network and the node characteristics when the social behavior is completed by using the series symbols.
6. The social behavior prediction method based on graph comparison learning according to claim 2, wherein the step of concatenating the network structure feature, the node influence feature, and the network inherent feature to generate an input layer feature specifically comprises:
respectively generating a feature matrix and an adjacent matrix of the positive sample subgraph and the anchor sample subgraph;
maximizing the consistency of the positive sample subgraph and the anchor point sample subgraph, deleting inconsistent nodes or edges in the two subgraphs, and representing the reserved nodes by a feature matrix; the two subgraphs are the positive sample subgraph and the anchor sample subgraph;
based on the feature matrix representation, the reserved nodes are projected to the two-layer perceptron to generate input layer features.
7. The social behavior prediction method based on graph comparison learning according to claim 2, wherein initializing the input layer features into a graph neural network predicts social behavior of a user, and specifically comprises:
Processing the input layer characteristics through a multi-layer network structure in the graph neural network to perform forward propagation;
for the characteristics of each layer of aggregation domain nodes in the multi-layer network structure, generating new characteristics of the nodes, and applying a ReLU activation function to the new characteristics;
calculating a plurality of category probability values by using a softmax activation function at the last layer of the multi-layer network structure, and determining a final predicted value;
and updating the gradient into the weight of the graph neural network through back propagation, and selecting the category corresponding to the final predicted value as the predicted social behavior of the user.
8. A graph-contrast learning-based social behavior prediction system, comprising:
the feature determining module is used for determining network structure features, node influence features and network inherent features according to the social influence network; the social influence network is established through social network data; the network structural features are obtained by pre-training a social influence network through graph comparison learning; the node influence characteristics are obtained by aggregating the characteristics and behavior states of neighbor nodes through a graph attention mechanism; the inherent network characteristics are obtained by connecting user profile characteristics, homogeneity characteristics in the social influence network and node characteristics when social behavior is completed in series;
The input layer characteristic generation module is used for splicing the network structure characteristic, the node influence characteristic and the network inherent characteristic to generate an input layer characteristic;
and the social behavior prediction module is used for initializing the input layer characteristics into the graphic neural network and predicting the social behavior of the user.
9. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the graph-contrast learning-based social behavior prediction method of any one of claims 1-7.
10. The electronic device of claim 9, wherein the memory is a non-transitory computer readable storage medium storing a computer program that when executed by a processor implements the graph-contrast learning-based social behavior prediction method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410145286.6A CN117670572B (en) | 2024-02-02 | 2024-02-02 | Social behavior prediction method, system and product based on graph comparison learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410145286.6A CN117670572B (en) | 2024-02-02 | 2024-02-02 | Social behavior prediction method, system and product based on graph comparison learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117670572A true CN117670572A (en) | 2024-03-08 |
| CN117670572B CN117670572B (en) | 2024-05-03 |
Family
ID=90068392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410145286.6A Active CN117670572B (en) | 2024-02-02 | 2024-02-02 | Social behavior prediction method, system and product based on graph comparison learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117670572B (en) |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020068831A1 (en) * | 2018-09-26 | 2020-04-02 | Visa International Service Association | Dynamic graph representation learning via attention networks |
| CN111931903A (en) * | 2020-07-09 | 2020-11-13 | 北京邮电大学 | Network alignment method based on double-layer graph attention neural network |
| CN112508085A (en) * | 2020-12-05 | 2021-03-16 | 西安电子科技大学 | Social network link prediction method based on perceptual neural network |
| CN112667920A (en) * | 2020-12-29 | 2021-04-16 | 复旦大学 | Text perception-based social influence prediction method, device and equipment |
| CN113050931A (en) * | 2021-03-15 | 2021-06-29 | 南京工业职业技术大学 | Symbolic network link prediction method based on graph attention machine mechanism |
| US20210248461A1 (en) * | 2020-02-11 | 2021-08-12 | Nec Laboratories America, Inc. | Graph enhanced attention network for explainable poi recommendation |
| CN113792937A (en) * | 2021-09-29 | 2021-12-14 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on graph neural network |
| CN114090902A (en) * | 2021-11-22 | 2022-02-25 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on heterogeneous network |
| CN114298851A (en) * | 2021-11-11 | 2022-04-08 | 国电南瑞科技股份有限公司 | Network user social behavior analysis method and device based on graph sign learning and storage medium |
| CN115168743A (en) * | 2022-07-27 | 2022-10-11 | 天津大学 | Crowd-sourcing body forwarding behavior prediction system based on heterogeneous graph neural network |
| CN115391710A (en) * | 2022-10-27 | 2022-11-25 | 暨南大学 | Social media account influence evaluation method and device based on graph neural network |
| CN115578211A (en) * | 2022-10-25 | 2023-01-06 | 南京工业职业技术大学 | Directed symbol network representation learning method and process with link prediction and node sequencing functions |
| CN115718831A (en) * | 2022-12-08 | 2023-02-28 | 人民网股份有限公司 | Social media robot detection method based on dynamic heterogeneous graph and related device |
| CN116304363A (en) * | 2023-01-19 | 2023-06-23 | 上海交通大学 | Social network influence maximum node searching method |
| CN116304866A (en) * | 2023-05-09 | 2023-06-23 | 海南大学 | Cross-network node classification method, equipment and medium based on graph comparison learning |
| CN116955846A (en) * | 2023-07-20 | 2023-10-27 | 重庆理工大学 | Cascade information propagation prediction method integrating theme characteristics and cross attention |
| CN116975778A (en) * | 2023-07-28 | 2023-10-31 | 郑州大学 | Social network information propagation influence prediction method based on information cascading |
| WO2024007599A1 (en) * | 2022-07-04 | 2024-01-11 | 上海淇玥信息技术有限公司 | Heterogeneous graph neural network-based method and apparatus for determining target service |
-
2024
- 2024-02-02 CN CN202410145286.6A patent/CN117670572B/en active Active
Patent Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020068831A1 (en) * | 2018-09-26 | 2020-04-02 | Visa International Service Association | Dynamic graph representation learning via attention networks |
| US20210248461A1 (en) * | 2020-02-11 | 2021-08-12 | Nec Laboratories America, Inc. | Graph enhanced attention network for explainable poi recommendation |
| CN111931903A (en) * | 2020-07-09 | 2020-11-13 | 北京邮电大学 | Network alignment method based on double-layer graph attention neural network |
| CN112508085A (en) * | 2020-12-05 | 2021-03-16 | 西安电子科技大学 | Social network link prediction method based on perceptual neural network |
| CN112667920A (en) * | 2020-12-29 | 2021-04-16 | 复旦大学 | Text perception-based social influence prediction method, device and equipment |
| CN113050931A (en) * | 2021-03-15 | 2021-06-29 | 南京工业职业技术大学 | Symbolic network link prediction method based on graph attention machine mechanism |
| CN113792937A (en) * | 2021-09-29 | 2021-12-14 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on graph neural network |
| CN114298851A (en) * | 2021-11-11 | 2022-04-08 | 国电南瑞科技股份有限公司 | Network user social behavior analysis method and device based on graph sign learning and storage medium |
| CN114090902A (en) * | 2021-11-22 | 2022-02-25 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on heterogeneous network |
| WO2024007599A1 (en) * | 2022-07-04 | 2024-01-11 | 上海淇玥信息技术有限公司 | Heterogeneous graph neural network-based method and apparatus for determining target service |
| CN115168743A (en) * | 2022-07-27 | 2022-10-11 | 天津大学 | Crowd-sourcing body forwarding behavior prediction system based on heterogeneous graph neural network |
| CN115578211A (en) * | 2022-10-25 | 2023-01-06 | 南京工业职业技术大学 | Directed symbol network representation learning method and process with link prediction and node sequencing functions |
| CN115391710A (en) * | 2022-10-27 | 2022-11-25 | 暨南大学 | Social media account influence evaluation method and device based on graph neural network |
| CN115718831A (en) * | 2022-12-08 | 2023-02-28 | 人民网股份有限公司 | Social media robot detection method based on dynamic heterogeneous graph and related device |
| CN116304363A (en) * | 2023-01-19 | 2023-06-23 | 上海交通大学 | Social network influence maximum node searching method |
| CN116304866A (en) * | 2023-05-09 | 2023-06-23 | 海南大学 | Cross-network node classification method, equipment and medium based on graph comparison learning |
| CN116955846A (en) * | 2023-07-20 | 2023-10-27 | 重庆理工大学 | Cascade information propagation prediction method integrating theme characteristics and cross attention |
| CN116975778A (en) * | 2023-07-28 | 2023-10-31 | 郑州大学 | Social network information propagation influence prediction method based on information cascading |
Non-Patent Citations (4)
| Title |
|---|
| HONGBO BO等: "Social Influence Prediction with Train and Test Time Augmentation for Graph Neural Networks", ARXIV:2104.11641V1, 23 April 2021 (2021-04-23), pages 1 - 8 * |
| 勤奋的IS\': "影响力最大化概述——《社会网络节点影响力分析研究》内容简述", pages 1 - 9, Retrieved from the Internet <URL:https://blog.csdn.net/lslslslslss/article/details/121831561> * |
| 鲍鹏;徐昊;: "基于图注意力时空神经网络的在线内容流行度预测", 模式识别与人工智能, no. 11, 15 November 2019 (2019-11-15), pages 1014 - 1021 * |
| 黎俊伟: "社交网络服务平台的用户行为预测研究", 中国博士学位论文全文数据库信息科技辑, no. 2, 15 February 2022 (2022-02-15), pages 139 - 23 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117670572B (en) | 2024-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| You et al. | Image-based appraisal of real estate properties | |
| CN111178458B (en) | Training of classification model, object classification method and device | |
| CN112287170B (en) | Short video classification method and device based on multi-mode joint learning | |
| CN116664719B (en) | Image redrawing model training method, image redrawing method and device | |
| CN112035743A (en) | Data recommendation method and device, computer equipment and storage medium | |
| CN113255798A (en) | Classification model training method, device, equipment and medium | |
| Khan et al. | Comparative analysis on Facebook post interaction using DNN, ELM and LSTM | |
| CN114282077A (en) | Session recommendation method and system based on session data | |
| CN116541608B (en) | House source recommendation method and device, electronic equipment and storage medium | |
| CN116628345B (en) | Content recommendation method and device, electronic equipment and storage medium | |
| CN113609337A (en) | Pre-training method, device, equipment and medium of graph neural network | |
| CN117670572B (en) | Social behavior prediction method, system and product based on graph comparison learning | |
| CN116522232A (en) | Document classification method, device, equipment and storage medium | |
| CN116361643A (en) | Model training method for realizing object recommendation, object recommendation method and related device | |
| CN114155388B (en) | Image recognition method and device, computer equipment and storage medium | |
| CN115564532A (en) | Training method and device of sequence recommendation model | |
| CN114493674A (en) | Advertisement click rate prediction model and method | |
| CN115169433A (en) | Knowledge graph classification method based on meta-learning and related equipment | |
| CN114519600A (en) | Graph neural network CTR estimation algorithm fusing adjacent node variances | |
| CN114168804A (en) | Similar information retrieval method and system based on heterogeneous subgraph neural network | |
| CN115329183A (en) | Data processing method, device, storage medium and equipment | |
| Chai et al. | An artistic analysis model based on sequence cartoon images for scratch | |
| CN111177493A (en) | Data processing method, device, server and storage medium | |
| CN113010772B (en) | Data processing method, related equipment and computer readable storage medium | |
| CN111598599A (en) | User characterization method and device, electronic equipment and computer readable medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |