CN117113240A - Dynamic network community discovery method, device, equipment and storage medium - Google Patents
Dynamic network community discovery method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN117113240A CN117113240A CN202311372307.XA CN202311372307A CN117113240A CN 117113240 A CN117113240 A CN 117113240A CN 202311372307 A CN202311372307 A CN 202311372307A CN 117113240 A CN117113240 A CN 117113240A
- Authority
- CN
- China
- Prior art keywords
- matrix
- dynamic
- community discovery
- graph
- encoder
- 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
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000003860 storage Methods 0.000 title claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims abstract description 221
- 230000008569 process Effects 0.000 claims abstract description 21
- 230000009977 dual effect Effects 0.000 claims abstract description 19
- 230000003993 interaction Effects 0.000 claims abstract description 16
- 230000006870 function Effects 0.000 claims description 75
- 238000009826 distribution Methods 0.000 claims description 45
- 230000004927 fusion Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000003064 k means clustering Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 4
- 238000004891 communication Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 101000926525 Homo sapiens eIF-2-alpha kinase GCN2 Proteins 0.000 description 1
- 101000862627 Homo sapiens eIF-2-alpha kinase activator GCN1 Proteins 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 102100034175 eIF-2-alpha kinase GCN2 Human genes 0.000 description 1
- 102100030495 eIF-2-alpha kinase activator GCN1 Human genes 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011273 social behavior Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of network information, and discloses a dynamic network community discovery method, a device, equipment and a storage medium, wherein the method comprises the following steps: obtaining a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process; constructing a dynamic network community discovery model based on a dynamic dual self-encoder; inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the previous moment into a dynamic network community discovery model to obtain a target loss function; and classifying the user communities based on the objective loss function to obtain dynamic network communities. Compared with the traditional community discovery method, the method divides the dynamic network communities based on the dynamic network community discovery model, so that the technical defect that the traditional community discovery method ignores the connection among various features generated by users in the social network and network dynamics is avoided, and the community structure in the dynamic network can be accurately discovered in real time.
Description
Technical Field
The present invention relates to the field of network information technologies, and in particular, to a method, an apparatus, a device, and a storage medium for dynamic network community discovery.
Background
The development of the social network promotes the communication of people and the propagation of information, and brings the complexity and the dynamic property of the network, and the community structure is an important characteristic of the social network, so that the user can be effectively recommended in a personalized way and the information propagation can be effectively monitored by carrying out community discovery on the social network.
With the continued development of virtual networks, the network structure of social networks is becoming more and more complex. To deal with more complex network structures, researchers have proposed some conventional community discovery methods. However, in the implementation of these traditional community discovery methods, the links between the various features that users produce in the social network are not considered; meanwhile, the traditional community discovery method also ignores the network dynamics, namely, network nodes and edges of the real world are changed continuously along with time, which leads to the change of a network topology structure and attributes on a fixed topology, thereby causing the change of the community structure. Based on the above reasons, the conventional community discovery method at present cannot accurately discover the community structure in the dynamic network in real time.
The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.
Disclosure of Invention
The invention mainly aims to provide a dynamic network community discovery method, a device, equipment and a storage medium, which aim to solve the technical problem that the community structure in a dynamic network cannot be accurately discovered in real time by the prior art method.
In order to achieve the above object, the present invention provides a dynamic network community discovery method, the method comprising the steps of:
obtaining a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process;
constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module;
inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the last moment into the dynamic network community discovery model to obtain a target loss function;
and classifying the user communities based on the target loss function to obtain dynamic network communities.
Optionally, the step of constructing a dynamic network community discovery model based on a dynamic dual self-encoder includes:
Constructing a depth self-encoder based on an attribute encoder and an attribute decoder, wherein the attribute encoder is used for encoding the node attribute matrix, the attribute decoder is used for decoding the node attribute matrix, and the structure of the attribute encoder and the structure of the attribute decoder are in a symmetrical relation;
constructing a graph attention variation self-encoder based on a graph encoder and a graph structure decoder, wherein the graph encoder is used for extracting the structure information of graph data in the dynamic social network data, and the graph structure decoder is used for reconstructing the structure information of the graph data;
and constructing a dynamic network community discovery model based on the dynamic dual self-encoder based on the depth self-encoder, the annotation meaning variation self-encoder and a time sequence feature modeling layer module.
Optionally, before the step of inputting the graph adjacency matrix, the node attribute matrix and the time sequence feature at the previous time into the dynamic network community discovery model to obtain the objective loss function, the method further includes:
the graph adjacent matrix at the previous moment, the node attribute matrix at the previous moment and the node potential vector at the previous moment are fused through a graph convolution network in the time sequence feature modeling layer module to obtain a time sequence fusion matrix, and the node potential vector is obtained after being learned based on an attribute encoder and a graph encoder;
And extracting the characteristics of the time sequence fusion matrix through a gating circulation unit to obtain the time sequence characteristics of the last time.
Optionally, the step of inputting the graph adjacency matrix, the node attribute matrix and the time sequence feature at the last moment into the dynamic network community discovery model to obtain a target loss function includes:
inputting the graph adjacent matrix, the node attribute matrix and the time sequence characteristic at the last moment into the graph encoder to obtain a normal distribution mean vector matrix and a normal distribution variance vector matrix;
inputting the node attribute matrix into the depth self-encoder to obtain a hidden attribute matrix and a reconstructed node attribute matrix;
and obtaining a target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstruction node attribute matrix.
Optionally, the step of obtaining the target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstructed node attribute matrix includes:
obtaining node attribute loss based on the reconstructed node attribute matrix and the node attribute matrix;
Obtaining graph adjacency loss based on the graph adjacency matrix and a reconstruction adjacency matrix, wherein the reconstruction adjacency matrix is obtained by inputting the graph adjacency matrix to the graph structure decoder for reconstruction;
obtaining KL divergence loss based on the normal distribution mean vector matrix and the normal distribution variance vector matrix;
and obtaining a target loss function according to the node attribute loss, the graph adjacency loss and the KL divergence loss.
Optionally, the step of classifying the user communities based on the objective loss function to obtain dynamic network communities includes:
optimizing the dynamic network community discovery model based on the target loss function to obtain an optimized dynamic network community discovery model;
and classifying the user communities through a k-means clustering algorithm to obtain dynamic network communities, and identifying the users based on the optimized dynamic network community discovery model to obtain the dynamic network communities.
Optionally, the step of optimizing the dynamic network community discovery model based on the objective loss function to obtain an optimized dynamic network community discovery model includes:
Training and learning the dynamic network community discovery model through the target loss function to obtain graph structure information, node attribute information and community potential characteristics;
and carrying out model optimization on the dynamic network community discovery model based on the graph structure information, the node attribute information and the community potential characteristics to obtain an optimized dynamic network community discovery model.
In addition, in order to achieve the above object, the present invention also proposes a dynamic web community discovery apparatus including:
the data acquisition module is used for acquiring a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated in the network interaction process of a user;
the model construction module is used for constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer;
the data input module is used for inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the last moment into the dynamic network community discovery model to obtain a target loss function;
And the category division module is used for carrying out category division on the user communities based on the target loss function to obtain dynamic network communities.
In addition, to achieve the above object, the present invention also proposes a dynamic network community discovery apparatus, the apparatus comprising: a memory, a processor, and a dynamic web community discovery program stored on the memory and executable on the processor, the dynamic web community discovery program configured to implement the steps of the dynamic web community discovery method as described above.
In addition, in order to achieve the above object, the present invention also proposes a storage medium having stored thereon a dynamic web community discovery program which, when executed by a processor, implements the steps of the dynamic web community discovery method as described above.
The method comprises the steps of obtaining a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process; constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module; inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the previous moment into a dynamic network community discovery model to obtain a target loss function; and classifying the user communities based on the objective loss function to obtain dynamic network communities. Compared with the traditional community discovery method, the method disclosed by the invention has the advantages that the dynamic network community discovery model is built based on the depth self-encoder, the graph meaning variation self-encoder and the time sequence feature modeling layer module, then the dynamic social network data generated in the network interaction process of the user is input into the dynamic network community discovery model to obtain the target loss function, and the user communities are classified based on the target loss function to obtain the dynamic network communities, so that the technical defect that the traditional community discovery method ignores the connection among various features generated in the social network by the user and the network dynamics is avoided, and the community structure in the dynamic network can be accurately discovered in real time.
Drawings
FIG. 1 is a schematic diagram of a dynamic network community discovery device of a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a flowchart of a dynamic network community discovery method according to a first embodiment of the present invention;
FIG. 3 is a flowchart illustrating a dynamic network community discovery method according to a second embodiment of the present invention;
FIG. 4 is a flowchart of a third embodiment of a dynamic network community discovery method of the present invention;
fig. 5 is a block diagram illustrating a dynamic network community discovery device according to a first embodiment of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Referring to fig. 1, fig. 1 is a schematic diagram of a dynamic network community discovery device of a hardware running environment according to an embodiment of the present invention.
As shown in fig. 1, the dynamic network community discovery apparatus may include: a processor 1001, such as a central processing unit (Central Processing Unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, a memory 1005. Wherein the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display, an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may further include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a Wireless interface (e.g., a Wireless-Fidelity (Wi-Fi) interface). The Memory 1005 may be a high-speed random access Memory (Random Access Memory, RAM) or a stable nonvolatile Memory (NVM), such as a disk Memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.
Those skilled in the art will appreciate that the structure shown in FIG. 1 is not limiting of the dynamic network community finding device and may include more or fewer components than shown, or may combine certain components, or may be a different arrangement of components.
As shown in fig. 1, an operating system, a network communication module, a user interface module, and a dynamic network community discovery program may be included in the memory 1005 as one type of storage medium.
In the dynamic web community discovery apparatus shown in fig. 1, the network interface 1004 is mainly used for data communication with a web server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the dynamic web community discovery apparatus of the present invention may be provided in the dynamic web community discovery apparatus, which invokes a dynamic web community discovery program stored in the memory 1005 through the processor 1001 and performs the dynamic web community discovery method provided by the embodiment of the present invention.
The embodiment of the invention provides a dynamic network community discovery method, referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the dynamic network community discovery method of the invention.
In this embodiment, the dynamic network community discovery method includes the following steps:
step S10: a graph adjacency matrix and a node attribute matrix are obtained based on dynamic social network data generated by a user in a network interaction process.
It should be noted that, the execution body of the method of the present embodiment may be a computing service device with functions of data processing, network communication and program running, for example, a mobile phone, a tablet computer, a personal computer, etc., or may be other electronic devices capable of implementing the same or similar functions, which is not limited in this embodiment. Various embodiments of the dynamic web community discovery method of the present invention will be described herein by taking a dynamic web community discovery device (hereinafter referred to as a discovery device) as an example.
It is appreciated that the dynamic social network data described above may be represented asWhereinIs a snapshot network of time step t, +.>And->A set of nodes and a set of edges of the snapshot network representing time step t. There may be one type of node and one type of edge in the network of the present embodiment, the number of nodes being +.>The number of sides is->. In particular, the dynamic social network data may be a DBLP-3 dataset, the number of dynamic nodes in the DBLP-3 dataset may be 4257, the number of dynamic edges may be 23540, the time step may be 10, the node feature dimension may be 100, and the number of communities may be 3; the node characteristic dimension may be 100, the community number may be 5, and the embodiment is not limited to this.
It should be appreciated that the graph adjacency matrix described above may beFor representing a snapshot network->Connection between nodes. Node->And->The presence of an edge between them can be expressed as +.>Otherwise->. The node attribute matrix may be +.>The node characteristics of (a) can be expressed as matrix +.>The ith row of the matrix represents node +.>The feature vector at time step t, if there are no node features, the matrix may be set to，/>Representing a unit vector.
Step S20: a dynamic network community discovery model based on a dynamic dual self-encoder is constructed, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence characteristic modeling layer module.
It should be noted that the depth self-encoder may include an attribute encoder and an attribute decoder, which are structurally symmetrical and may be used to encode and decode attribute information. In this embodiment, the depth self-encoder may be used to extract attribute information, and the attribute hidden representation extracted by the middle layer may be transferred to the attention-variant self-encoder, and the learned attribute information may be supplemented to the attention-variant self-encoder.
It should be appreciated that the above-described graph meaning variations from encoder may include a graph encoder and a graph structure decoder. The graph structure decoder uses the graph attention network to extract the structure information, and the graph structure decoder uses the vector inner product to reconstruct the adjacent matrix, and learns the potential representation of the structure information by minimizing the graph structure reconstruction loss.
It can be understood that the above-mentioned time sequence feature modeling layer module can be used to model the changes of the dynamic isomorphic network structure information and the attribute information, firstly, the two-layer graph convolution network is used to fuse the information of the network structure, the attribute feature vector and the node potential vector at the last moment, and then the dependency of the learning snapshot network is modeled through the optimized GRU (Gate Recurrent Unit, gating circulation unit), and the time sequence feature of the dynamic network is learned.
Step S30: and inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the last moment into the dynamic network community discovery model to obtain a target loss function.
It should be noted that, the objective loss function may be a function for measuring a difference between a predicted result and an actual tag in machine learning or deep learning. The method is a core index for model training and is used for measuring the performance of the model and guiding the direction of parameter optimization. More specifically, the objective loss function may include a mean square error loss function, a cross entropy loss function, a log loss function, a negative log likelihood loss function, or other types of loss functions, which the present embodiment is not limited to.
Step S40: and classifying the user communities based on the target loss function to obtain dynamic network communities.
The dynamic web community may be a web community to which the social behavior of the user in the dynamic social network belongs. Wherein, the network nodes and edges of the dynamic social network are changed continuously along with the time change.
In a specific implementation, the dynamic network community discovery model may be optimized by providing an optimization direction based on the objective loss function, so that the user communities to which the user belongs may be classified based on the optimized dynamic network community discovery model, and a dynamic network community may be obtained.
The embodiment obtains a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process; constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module; inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the previous moment into a dynamic network community discovery model to obtain a target loss function; and classifying the user communities based on the objective loss function to obtain dynamic network communities. Compared with the traditional community discovery method, the method of the embodiment builds the dynamic network community discovery model based on the depth self-encoder, the graph meaning variation self-encoder and the time sequence feature modeling layer module, then inputs dynamic social network data generated by users in the network interaction process into the dynamic network community discovery model to obtain the target loss function, and classifies the user communities based on the target loss function to obtain the dynamic network communities, so that the technical defect that the traditional community discovery method ignores the connection among various features generated by the users in the social network and network dynamics is avoided, and further the community structure in the dynamic network can be accurately discovered in real time.
Referring to fig. 3, fig. 3 is a flowchart illustrating a dynamic network community discovery method according to a second embodiment of the present invention.
Based on the above-mentioned first embodiment, in this embodiment, in order to construct a dynamic network community discovery model of a dynamic dual self-encoder, so as to implement reconstruction of dynamic social network data generated by a user in a network interaction process, the step S20 may include:
step S201: and constructing a depth self-encoder based on an attribute encoder and an attribute decoder, wherein the attribute encoder is used for encoding the node attribute matrix, the attribute decoder is used for decoding the node attribute matrix, and the structure of the attribute encoder and the structure of the attribute decoder are in a symmetrical relation.
In a specific implementation, the attribute decoder may reconstruct the node attribute matrix based on a fully connected neural network, so as to implement a decoding action on the node attribute matrix, and may specifically be expressed as:
wherein, in the above formulaNon-linear activation function representing fully connected layer, < ->And->Respectively representing the weight matrix and the offset of the first layer of the attribute decoder, and obtaining a reconstructed node attribute matrix after the low-dimensional hidden representation of the node attribute information is decoded by the attribute decoder >。
Step S202: and constructing a graph attention variation self-encoder based on a graph encoder and a graph structure decoder, wherein the graph encoder is used for extracting the structure information of the graph data in the dynamic social network data, and the graph structure decoder is used for reconstructing the structure information of the graph data.
In a specific implementation, the graph encoder can learn the distribution represented by the nodes by adopting a two-layer graph annotation force network and taking the output of time sequence feature modeling at the previous moment, a graph adjacent matrix at the current moment and a node feature matrix as inputs, thereby extracting the structural information of the graph data. The above-described graph structure decoder reconstructs the graph structure by calculating the probability of edge-to-edge between any two nodes in the graph in the form of a vector inner product.
Step S203: and constructing a dynamic network community discovery model based on the dynamic dual self-encoder based on the depth self-encoder, the annotation meaning variation self-encoder and a time sequence feature modeling layer module.
Step S204: and fusing the graph adjacent matrix at the previous moment, the node attribute matrix at the previous moment and the node potential vector at the previous moment through a graph rolling network in the time sequence feature modeling layer module to obtain a time sequence fusion matrix, wherein the node potential vector is obtained after being learned based on an attribute encoder and a graph encoder.
In a specific implementation, the above-described timing fusion matrix can be represented by the following formula:
where GCN1 may represent a first layer of graph roll-up network, GCN2 may represent a second layer of graph roll-up network,graph adjacency matrix capable of representing the previous time, and +.>Node attribute matrix capable of representing the previous time>The node potential vector of the previous moment can be represented,/->Representing matrix join operations, +.>Is a multi-layer perceptron network, independently operates on each node, from +.>And->Features are extracted.
Step S205: and extracting the characteristics of the time sequence fusion matrix through a gating circulation unit to obtain the time sequence characteristics of the last time.
Based on the first embodiment, in this embodiment, in order to obtain a more accurate target loss function, so that the performance of the model and the direction of guiding parameter optimization can be measured, the step S30 may include:
step S301: and inputting the graph adjacent matrix, the node attribute matrix and the time sequence characteristic at the last moment into the graph encoder to obtain a normal distribution mean vector matrix and a normal distribution variance vector matrix.
It should be understood that the above-described normal distribution mean vector matrix and normal distribution variance vector matrix can be expressed by the following formulas:
Wherein,a mean vector matrix representing the normal distribution, < >>A variance vector matrix representing the normal distribution, < >>Representing the above-mentioned last time sequence feature, +.>Representing matrix join operations, +.>Representing deep neural networks which operate independently on each node and from +.>And extracting the characteristics. It should be noted that the first layer diagramAttention network->Is shared. The structural potential vector representation of the node is a slave vector distribution +.>Since the training model uses gradient descent, the sampling operation has no way to counter-propagate, requiring the use of re-parameterization to avoid this problem. More specifically, the process of re-parameterization can be expressed as:
wherein,representing the linear transformation variable +.>，/>Representing the hadamard product of the matrix.
Step S302: and inputting the node attribute matrix into the depth self-encoder to obtain a hidden attribute matrix and a reconstructed node attribute matrix.
In a specific implementation, the hidden attribute matrix described above may be expressed based on the following formula:
wherein,layer I hidden attribute matrix representing time step t in depth self-encoder,/for the encoder>Non-linear activation function representing fully connected layer, < - >And->The weight matrix and the offset of the first layer of the attribute encoder are represented. In particular, the input of the first layer is the original node feature matrix, i.e. +.>。
Step S303: and obtaining a target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstruction node attribute matrix.
In a specific implementation, the average vector matrix of normal distribution, the variance vector matrix of normal distribution, the hidden attribute matrix and the attribute matrix of reconstruction node are assigned with weights, and the average vector matrix of normal distribution, the variance vector matrix of normal distribution, the hidden attribute matrix and the attribute matrix of reconstruction node after being assigned with the weights are subjected to superposition operation, so as to obtain the target loss function.
In the embodiment, a depth self-encoder is constructed based on an attribute encoder and an attribute decoder, wherein the attribute encoder is used for encoding a node attribute matrix, the attribute decoder is used for decoding the node attribute matrix, and the structures of the attribute encoder and the attribute decoder are in a symmetrical relation; constructing a graph attention variation self-encoder based on a graph encoder and a graph structure decoder, wherein the graph encoder is used for extracting structure information of graph data in dynamic social network data, and the graph structure decoder is used for reconstructing the structure information of the graph data; constructing a dynamic network community discovery model based on a dynamic dual self-encoder based on a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module; the graph adjacent matrix at the previous moment, the node attribute matrix at the previous moment and the node potential vector at the previous moment are fused through a graph rolling network in the time sequence feature modeling layer module to obtain a time sequence fusion matrix, and the node potential vector is obtained after being learned based on an attribute encoder and a graph encoder; extracting features of the time sequence fusion matrix through a gate control circulation unit to obtain the time sequence features of the last time; inputting the graph adjacent matrix, the node attribute matrix and the time sequence characteristic at the last moment into a graph encoder to obtain a normal distribution mean vector matrix and a normal distribution variance vector matrix; inputting the node attribute matrix into a depth self-encoder to obtain a hidden attribute matrix and a reconstructed node attribute matrix; and obtaining a target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstructed node attribute matrix. Compared with the traditional network community discovery method, the method of the embodiment realizes the reconstruction of dynamic social network data generated by a user in the network interaction process by constructing the dynamic network community discovery model of the dynamic dual self-encoder, so that the target loss function can be accurately obtained, and further, the node characteristics which are closer to the real condition of the community can be obtained based on the target loss function.
Referring to fig. 4, fig. 4 is a flowchart illustrating a third embodiment of the dynamic network community discovery method of the present invention.
Based on the foregoing embodiments, in this embodiment, in order to obtain the objective loss function more accurately, so as to obtain the node feature closer to the real situation of the community, the step S303 may include:
step S3031: and obtaining node attribute loss based on the reconstructed node attribute matrix and the node attribute matrix.
In a specific implementation, MSE (Mean Square Error, mean square error function) may be used to measure all time steps tAnd->And the difference is obtained, so that the node attribute loss is obtained. More specifically, it can be expressed by the following formula:
wherein,representing the above node attribute loss,/-, for>Representing the node attribute matrix,/->Representing the above reconstructed node attribute matrix, and N represents the number of nodes.
Step S3032: and obtaining graph adjacency loss based on the graph adjacency matrix and a reconstruction adjacency matrix, wherein the reconstruction adjacency matrix is obtained by inputting the graph adjacency matrix to the graph structure decoder for reconstruction.
Step S3033: and obtaining the KL divergence loss based on the normal distribution mean vector matrix and the normal distribution variance vector matrix.
In a specific implementation, the graph adjacency loss and KL divergence loss described above can be represented by the following formulas:
wherein,representing cross entropy loss of the graph adjacency matrix and the reconstructed adjacency matrix, +.>A mean vector matrix representing the normal distribution, < >>A variance vector matrix representing the normal distribution, < >>Representing unit vector +_>Representing a normal distribution.
Step S3034: and obtaining a target loss function according to the node attribute loss, the graph adjacency loss and the KL divergence loss.
In a specific implementation, the above equation L1 and the above equation L2 may be added to obtain the above objective loss function.
Based on the above embodiments, in this embodiment, in order to more accurately classify the user communities, the step S40 may include:
step S401: and optimizing the dynamic network community discovery model based on the target loss function to obtain an optimized dynamic network community discovery model.
In a specific implementation, the model parameters of the dynamic network community discovery model may be adjusted based on the objective loss function, and the model optimization process may be implemented by minimizing the value of the objective loss function.
Step S402: and classifying the user communities through a k-means clustering algorithm to obtain dynamic network communities, and identifying the users based on the optimized dynamic network community discovery model to obtain the dynamic network communities.
Based on the above embodiments, in this embodiment, in order to purposefully perform training learning on the dynamic web community discovery model, thereby improving the model performance of the dynamic web community discovery model, the step S40 may include:
step S4011: and training and learning the dynamic network community discovery model through the target loss function to obtain graph structure information, node attribute information and community potential characteristics.
Step S4012: and carrying out model optimization on the dynamic network community discovery model based on the graph structure information, the node attribute information and the community potential characteristics to obtain an optimized dynamic network community discovery model.
The embodiment obtains node attribute loss based on the reconstructed node attribute matrix and the node attribute matrix; obtaining graph adjacency loss based on the graph adjacency matrix and a reconstructed graph adjacency matrix, wherein the reconstructed graph adjacency matrix is obtained by inputting the graph adjacency matrix to a graph structure decoder for reconstruction; obtaining KL divergence loss based on the normal distribution mean vector matrix and the normal distribution variance vector matrix; obtaining a target loss function according to the node attribute loss, the graph adjacency loss and the KL divergence loss; training and learning the dynamic network community discovery model through the target loss function to obtain graph structure information, node attribute information and community potential characteristics; model optimization is carried out on the dynamic network community discovery model based on the graph structure information, the node attribute information and the community potential characteristics, and an optimized dynamic network community discovery model is obtained; and classifying the user communities by a k-means clustering algorithm to obtain dynamic network communities, and identifying the users by the user communities based on the optimized dynamic network community discovery model to obtain the dynamic network communities. Compared with the traditional network community discovery method, the method provided by the embodiment of the invention has the advantages that the k-means clustering algorithm is used for classifying the user communities obtained after the user is identified based on the optimized dynamic network community discovery model, so that the community structure in the dynamic network can be discovered more accurately.
In addition, the embodiment of the invention also provides a storage medium, wherein the storage medium is stored with a dynamic network community discovery program, and the dynamic network community discovery program realizes the steps of the dynamic network community discovery method when being executed by a processor.
Referring to fig. 5, fig. 5 is a block diagram illustrating a first embodiment of a dynamic network community discovery device according to the present invention.
As shown in fig. 5, the dynamic network community discovery device provided by the embodiment of the invention includes:
the data acquisition module 501 is configured to obtain a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process;
the model construction module 502 is configured to construct a dynamic network community discovery model based on a dynamic dual self-encoder, where the dynamic network community discovery model includes a depth self-encoder, a graph annotation meaning variation self-encoder, and a time sequence feature modeling layer;
the data input module 503 is configured to input the graph adjacency matrix, the node attribute matrix, and the time sequence feature at the previous time into the dynamic network community discovery model, so as to obtain a target loss function;
and the category classification module 504 is configured to perform category classification on the user community based on the objective loss function, so as to obtain a dynamic network community.
The embodiment obtains a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process; constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module; inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the previous moment into a dynamic network community discovery model to obtain a target loss function; and classifying the user communities based on the objective loss function to obtain dynamic network communities. Compared with the traditional community discovery method, the method of the embodiment builds the dynamic network community discovery model based on the depth self-encoder, the graph meaning variation self-encoder and the time sequence feature modeling layer module, then inputs dynamic social network data generated by users in the network interaction process into the dynamic network community discovery model to obtain the target loss function, and classifies the user communities based on the target loss function to obtain the dynamic network communities, so that the technical defect that the traditional community discovery method ignores the connection among various features generated by the users in the social network and network dynamics is avoided, and further the community structure in the dynamic network can be accurately discovered in real time.
Based on the first embodiment of the dynamic web community discovery apparatus of the present invention described above, a second embodiment of the dynamic web community discovery apparatus of the present invention is presented.
In this embodiment, the model building module 502 is further configured to build a depth self-encoder based on an attribute encoder and an attribute decoder, where the attribute encoder is configured to encode the node attribute matrix, the attribute decoder is configured to decode the node attribute matrix, and the structure of the attribute encoder and the structure of the attribute decoder are in a symmetrical relationship; constructing a graph attention variation self-encoder based on a graph encoder and a graph structure decoder, wherein the graph encoder is used for extracting the structure information of graph data in the dynamic social network data, and the graph structure decoder is used for reconstructing the structure information of the graph data; and constructing a dynamic network community discovery model based on the dynamic dual self-encoder based on the depth self-encoder, the annotation meaning variation self-encoder and a time sequence feature modeling layer module.
Further, the data input module 503 is further configured to fuse, through a graph convolution network in the timing characteristic modeling layer module, a graph adjacency matrix at a previous time, a node attribute matrix at a previous time, and a node potential vector at a previous time, to obtain a timing fusion matrix, where the node potential vector is obtained after learning based on an attribute encoder and a graph encoder; and extracting the characteristics of the time sequence fusion matrix through a gating circulation unit to obtain the time sequence characteristics of the last time.
Further, the data input module 503 is further configured to input the graph adjacency matrix, the node attribute matrix, and the time sequence feature at the previous time to the graph encoder, so as to obtain a normal distributed mean vector matrix and a normal distributed variance vector matrix; inputting the node attribute matrix into the depth self-encoder to obtain a hidden attribute matrix and a reconstructed node attribute matrix; and obtaining a target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstruction node attribute matrix.
Further, the data input module 503 is further configured to obtain a node attribute loss based on the reconstructed node attribute matrix and the node attribute matrix; obtaining graph adjacency loss based on the graph adjacency matrix and a reconstruction adjacency matrix, wherein the reconstruction adjacency matrix is obtained by inputting the graph adjacency matrix to the graph structure decoder for reconstruction; obtaining KL divergence loss based on the normal distribution mean vector matrix and the normal distribution variance vector matrix; and obtaining a target loss function according to the node attribute loss, the graph adjacency loss and the KL divergence loss.
Further, the category classification module 504 is further configured to optimize the dynamic network community discovery model based on the objective loss function, so as to obtain an optimized dynamic network community discovery model; and classifying the user communities through a k-means clustering algorithm to obtain dynamic network communities, and identifying the users based on the optimized dynamic network community discovery model to obtain the dynamic network communities.
Further, the category classification module 504 is further configured to perform training learning on the dynamic network community discovery model through the objective loss function to obtain graph structure information, node attribute information and community potential characteristics; and carrying out model optimization on the dynamic network community discovery model based on the graph structure information, the node attribute information and the community potential characteristics to obtain an optimized dynamic network community discovery model.
Other embodiments or specific implementation manners of the dynamic network community discovery device of the present invention may refer to the above method embodiments, and are not described herein.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system 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 system. 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 system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of embodiments, it will be clear to a person skilled in the art that the above embodiment method may be implemented by means of software plus a necessary general hardware platform, but may of course also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. read-only memory/random-access memory, magnetic disk, optical disk), comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (10)
1. A method of dynamic web community discovery, the method comprising the steps of:
obtaining a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated by a user in a network interaction process;
constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer module;
inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the last moment into the dynamic network community discovery model to obtain a target loss function;
and classifying the user communities based on the target loss function to obtain dynamic network communities.
2. The dynamic web community discovery method of claim 1, wherein the step of constructing a dynamic web community discovery model based on a dynamic dual self-encoder comprises:
constructing a depth self-encoder based on an attribute encoder and an attribute decoder, wherein the attribute encoder is used for encoding the node attribute matrix, the attribute decoder is used for decoding the node attribute matrix, and the structure of the attribute encoder and the structure of the attribute decoder are in a symmetrical relation;
Constructing a graph attention variation self-encoder based on a graph encoder and a graph structure decoder, wherein the graph encoder is used for extracting the structure information of graph data in the dynamic social network data, and the graph structure decoder is used for reconstructing the structure information of the graph data;
and constructing a dynamic network community discovery model based on the dynamic dual self-encoder based on the depth self-encoder, the annotation meaning variation self-encoder and a time sequence feature modeling layer module.
3. The method for dynamic web community discovery according to claim 1, wherein before the step of inputting the graph adjacency matrix, the node attribute matrix, and the last time sequential feature into the dynamic web community discovery model to obtain the objective loss function, the method further comprises:
the graph adjacent matrix at the previous moment, the node attribute matrix at the previous moment and the node potential vector at the previous moment are fused through a graph convolution network in the time sequence feature modeling layer module to obtain a time sequence fusion matrix, and the node potential vector is obtained after being learned based on an attribute encoder and a graph encoder;
and extracting the characteristics of the time sequence fusion matrix through a gating circulation unit to obtain the time sequence characteristics of the last time.
4. The dynamic web community discovery method of claim 2, wherein the step of inputting the graph adjacency matrix, the node attribute matrix, and the last time sequential feature into the dynamic web community discovery model to obtain an objective loss function comprises:
inputting the graph adjacent matrix, the node attribute matrix and the time sequence characteristic at the last moment into the graph encoder to obtain a normal distribution mean vector matrix and a normal distribution variance vector matrix;
inputting the node attribute matrix into the depth self-encoder to obtain a hidden attribute matrix and a reconstructed node attribute matrix;
and obtaining a target loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix and the reconstruction node attribute matrix.
5. The dynamic network community finding method as claimed in claim 4, wherein the step of obtaining the objective loss function based on the normal distribution mean vector matrix, the normal distribution variance vector matrix, the hidden attribute matrix, and the reconstructed node attribute matrix comprises:
obtaining node attribute loss based on the reconstructed node attribute matrix and the node attribute matrix;
Obtaining graph adjacency loss based on the graph adjacency matrix and a reconstruction adjacency matrix, wherein the reconstruction adjacency matrix is obtained by inputting the graph adjacency matrix to the graph structure decoder for reconstruction;
obtaining KL divergence loss based on the normal distribution mean vector matrix and the normal distribution variance vector matrix;
and obtaining a target loss function according to the node attribute loss, the graph adjacency loss and the KL divergence loss.
6. The method for discovering dynamic network communities as in claim 1, wherein the step of classifying the user communities based on the objective loss function to obtain dynamic network communities includes:
optimizing the dynamic network community discovery model based on the target loss function to obtain an optimized dynamic network community discovery model;
and classifying the user communities through a k-means clustering algorithm to obtain dynamic network communities, and identifying the users based on the optimized dynamic network community discovery model to obtain the dynamic network communities.
7. The dynamic web community discovery method of claim 6, wherein the step of optimizing the dynamic web community discovery model based on the objective loss function to obtain an optimized dynamic web community discovery model comprises:
Training and learning the dynamic network community discovery model through the target loss function to obtain graph structure information, node attribute information and community potential characteristics;
and carrying out model optimization on the dynamic network community discovery model based on the graph structure information, the node attribute information and the community potential characteristics to obtain an optimized dynamic network community discovery model.
8. A dynamic web community discovery apparatus, the dynamic web community discovery apparatus comprising:
the data acquisition module is used for acquiring a graph adjacency matrix and a node attribute matrix based on dynamic social network data generated in the network interaction process of a user;
the model construction module is used for constructing a dynamic network community discovery model based on a dynamic dual self-encoder, wherein the dynamic network community discovery model comprises a depth self-encoder, a graph annotation meaning variation self-encoder and a time sequence feature modeling layer;
the data input module is used for inputting the graph adjacency matrix, the node attribute matrix and the time sequence characteristic at the last moment into the dynamic network community discovery model to obtain a target loss function;
and the category division module is used for carrying out category division on the user communities based on the target loss function to obtain dynamic network communities.
9. A dynamic web community discovery device, the device comprising: a memory, a processor, and a dynamic web community discovery program stored on the memory and executable on the processor, the dynamic web community discovery program configured to implement the steps of the dynamic web community discovery method of any one of claims 1 to 7.
10. A storage medium having stored thereon a dynamic web community discovery program which when executed by a processor performs the steps of the dynamic web community discovery method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311372307.XA CN117113240B (en) | 2023-10-23 | 2023-10-23 | Dynamic network community discovery method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311372307.XA CN117113240B (en) | 2023-10-23 | 2023-10-23 | Dynamic network community discovery method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117113240A true CN117113240A (en) | 2023-11-24 |
| CN117113240B CN117113240B (en) | 2024-03-26 |
Family
ID=88811352
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311372307.XA Active CN117113240B (en) | 2023-10-23 | 2023-10-23 | Dynamic network community discovery method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117113240B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112084328A (en) * | 2020-07-29 | 2020-12-15 | 浙江工业大学 | Scientific and technological thesis clustering analysis method based on variational graph self-encoder and K-Means |
| CN112905656A (en) * | 2021-01-29 | 2021-06-04 | 重庆理工大学 | Dynamic community discovery system fusing time sequence network |
| CN113407784A (en) * | 2021-05-28 | 2021-09-17 | 桂林电子科技大学 | Social network-based community dividing method, system and storage medium |
| CN114020999A (en) * | 2021-10-20 | 2022-02-08 | 山西大学 | Community structure detection method and system for movie social network |
| WO2022166361A1 (en) * | 2021-02-04 | 2022-08-11 | 浙江师范大学 | Deep clustering method and system based on cross-modal fusion |
| CN116595479A (en) * | 2023-05-06 | 2023-08-15 | 云南大学 | Community discovery method, system, equipment and medium based on graph double self-encoder |
-
2023
- 2023-10-23 CN CN202311372307.XA patent/CN117113240B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112084328A (en) * | 2020-07-29 | 2020-12-15 | 浙江工业大学 | Scientific and technological thesis clustering analysis method based on variational graph self-encoder and K-Means |
| CN112905656A (en) * | 2021-01-29 | 2021-06-04 | 重庆理工大学 | Dynamic community discovery system fusing time sequence network |
| WO2022166361A1 (en) * | 2021-02-04 | 2022-08-11 | 浙江师范大学 | Deep clustering method and system based on cross-modal fusion |
| CN113407784A (en) * | 2021-05-28 | 2021-09-17 | 桂林电子科技大学 | Social network-based community dividing method, system and storage medium |
| CN114020999A (en) * | 2021-10-20 | 2022-02-08 | 山西大学 | Community structure detection method and system for movie social network |
| CN116595479A (en) * | 2023-05-06 | 2023-08-15 | 云南大学 | Community discovery method, system, equipment and medium based on graph double self-encoder |
Non-Patent Citations (2)
| Title |
|---|
| YUANXIN LI ET AL.: "Maximizing the Influence of Social Networks Based on Graph Attention Networks", 《2023 3RD ASIA-PACIFIC CONFERENCE ON COMMUNICATIONS TECHNOLOGY AND COMPUTER SCIENCE (ACCTCS)》, pages 396 - 403 * |
| 黄瑞阳;吴奇;朱宇航;: "基于联合矩阵分解的动态异质网络社区发现方法", 计算机应用研究, no. 10, 31 October 2017 (2017-10-31), pages 115 - 118 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117113240B (en) | 2024-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10387768B2 (en) | Enhanced restricted boltzmann machine with prognosibility regularization for prognostics and health assessment | |
| CN111695415B (en) | Image recognition method and related equipment | |
| CA3061667C (en) | Systems and methods for assessing item compatibility | |
| WO2022105117A1 (en) | Method and device for image quality assessment, computer device, and storage medium | |
| CN113628059B (en) | Associated user identification method and device based on multi-layer diagram attention network | |
| Steingrimsson et al. | Deep learning for survival outcomes | |
| Tomy et al. | Estimating the state of epidemics spreading with graph neural networks | |
| CN113065974A (en) | Link prediction method based on dynamic network representation learning | |
| CN116310667A (en) | Self-supervision visual characterization learning method combining contrast loss and reconstruction loss | |
| CN115271980A (en) | Risk value prediction method and device, computer equipment and storage medium | |
| CN112667920A (en) | Text perception-based social influence prediction method, device and equipment | |
| Deng et al. | Network Intrusion Detection Based on Sparse Autoencoder and IGA‐BP Network | |
| Devi et al. | Smoothing approach to alleviate the meager rating problem in collaborative recommender systems | |
| CN114896977A (en) | Dynamic evaluation method for entity service trust value of Internet of things | |
| CN113239266B (en) | Personalized recommendation method and system based on local matrix decomposition | |
| Karupusamy et al. | Effective energy usage and data compression approach using data mining algorithms for IoT data | |
| CN117113240B (en) | Dynamic network community discovery method, device, equipment and storage medium | |
| Zhu et al. | A hybrid model for nonlinear regression with missing data using quasilinear kernel | |
| CN113065321B (en) | User behavior prediction method and system based on LSTM model and hypergraph | |
| Wang et al. | Enhanced by mobility? Effect of users’ mobility on information diffusion in coupled online social networks | |
| CN115545833A (en) | Recommendation method and system based on user social information | |
| Kefeng et al. | CNN Based No‐Reference HDR Image Quality Assessment | |
| Sharma et al. | Comparative analysis of different algorithms in link prediction on social networks | |
| CN113626685A (en) | Propagation uncertainty-oriented rumor detection method and device | |
| Jia et al. | Dynamic influence prediction of social network based on partial autoregression single index model |
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 |