CN116307022A - Public opinion hotspot information prediction method and system - Google Patents
Public opinion hotspot information prediction method and system Download PDFInfo
- Publication number
- CN116307022A CN116307022A CN202211244827.8A CN202211244827A CN116307022A CN 116307022 A CN116307022 A CN 116307022A CN 202211244827 A CN202211244827 A CN 202211244827A CN 116307022 A CN116307022 A CN 116307022A
- Authority
- CN
- China
- Prior art keywords
- forwarding
- information
- dynamic routing
- cac
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000004220 aggregation Methods 0.000 claims abstract description 16
- 230000002776 aggregation Effects 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 230000010365 information processing Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000002775 capsule Substances 0.000 claims description 5
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 4
- 238000009792 diffusion process Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000011160 research Methods 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 2
- 230000004931 aggregating effect Effects 0.000 claims 1
- 238000013135 deep learning Methods 0.000 abstract description 3
- 238000003672 processing method Methods 0.000 abstract 1
- 238000013527 convolutional neural network Methods 0.000 description 16
- 238000010586 diagram Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 230000001609 comparable effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011551 log transformation method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000001550 time effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a system for predicting public opinion hotspot information, which are applied to the field of network information security. The system reasonably applies a deep learning method to model the forwarding process of the hot spot information, fully considers the structure and time information, and realizes the end-to-end prediction. The system comprises: converting the forwarding process of the hot spot information into a graph structure; extracting forwarding graph information by a cascade attention convolutional network (CAC); dynamic Routing-AT aggregation CAC extracted information; a gate control loop unit (GRU) processes time information; finally, predicting by using a multi-layer perceptron (MLP); and (5) model testing. The invention provides a brand-new CAC-G system aiming at complex data prediction of a social platform for the field, and also provides a brand-new forwarding graph processing method and an aggregation mode, so that the problems of low utilization rate, low prediction efficiency and the like of noisy social media data in the field are solved, and the requirements of dynamic capture and rapid and accurate prediction of public opinion hotspot information are met.
Description
Technical Field
The invention belongs to the field of network information security, and further relates to a public opinion hotspot information prediction method and system
Background
The network public opinion uses a network as a carrier and events as cores, and is a collection of vast network citizens emotion, attitude, opinion, expression of views, transmission and interaction and subsequent influence. The method has subjectivity of vast netizens, and is directly released on the Internet in various forms without media verification and packaging. Therefore, the network public opinion also has the characteristics of freedom, interactivity, pluripotency, deviation, burstiness and the like. The network public opinion has become an important factor for influencing the continuous and orderly development of society and maintaining the harmony and stability of the society. How to guide the network public opinion information in a factor, improve the analysis capability of the network public opinion information, timely and accurately master the social public opinion dynamics, and actively guide the social public opinion, is a serious subject and serious challenge faced by the network as an emerging medium.
The current public opinion hotspot information analysis method comprises the following steps: firstly, carrying out cluster analysis on the captured public opinion information; secondly, the digital expression form of public opinion information such as the number of participants of the public opinion information, the time distribution characteristics of the information reply and the like is analyzed; thirdly, analysis is performed by applying deep learning algorithm modeling from the forming and developing processes of hot spot public opinion. The first class and the second class can judge hot spot information, but have obvious hysteresis; the third type of analysis flow is reasonable and is the main public opinion hotspot prediction method at present, and the representative models mainly include DeepCas, AECasN and the like. The deep Cas is a first end-to-end model based on deep learning, the model obtains a set of node sequences in a random walk mode, the model mainly utilizes structural information and node information of a cascade graph, and the model predicts the growth scale of cascade by bidirectional GRU and an attention mechanism. AECasN proposes dividing the information cascade network into different layers and obtaining a preliminary representation of the cascade network, then multiplying the representation vector by a discrete vector of the time attenuation effect, inputting the discrete vector into an automatic encoder for learning, and outputting the cascade growth scale. Although the previous models are mature, the models cannot fully perform due to the fact that the data of various social platforms are too noisy. It follows that there is a continuing need to explore for more accurate and efficient ways.
Disclosure of Invention
In order to solve the problems, the invention provides a brand new CAC-G system for predicting complex data of a social platform, which has the following basic ideas: firstly, converting the forwarding process of the message into a graph form according to past experience, and continuously dividing subgraphs according to the time transition; it is then proposed to learn the node representations in the subgraph using the CAC model, then aggregate the representations of all nodes with Dynamic Routing-AT and process the hidden time information by means of the GRU, finally send into the MLP output cascade growth size. In order to verify the effectiveness of the invention, experiments are carried out on 6 sub-data sets in two different scenes of the New wave microblog and the paper citation, and the results show that the CAC-G system provided by the invention can give better predictions. The method comprises the following specific steps:
step 1: the method comprises the steps of continuously dividing a converted graph structure into forwarding subgraphs according to the time lapse, learning subgraph features through a CAC model, wherein the CAC model organically combines GAT and CNN, sends node features and edge features into GAT for learning, inputs the node features into CNN, and finally performs splicing operation on the outputs of the GAT and CNN to obtain output features of the forwarding subgraphs;
step 2: combining a Self-attribute weight calculation mode with a Dynamic Routing algorithm in a capsule network to provide Dynamic Routing-AT, and sending forwarding sub-graph features output by CAC into the Dynamic Routing-AT for aggregation to obtain vector representation of the sub-graph;
step 3: the invention uses GRU to process time information, the subgraph is divided continuously along with the time, new nodes are added into the diffusion process continuously, the time information is hidden in the subgraph sequence processed by Dynamic Routing-AT, and the prediction is carried out after the time information is processed;
step 4: predicting the forwarding cascade data scale through the constructed model;
testing the model, and putting part of data on the model to test by using the parameters learned during training to obtain a test result;
furthermore, in the step 1, a model CAC for complex data processing of the social platform is provided innovatively, and GAT and CNN are organically combined. The GAT with multiple head attentions is used in the CAC, the node characteristics and the edge characteristics of the forwarding subgraph are input into the GAT, meanwhile, the node characteristics are input into the CNN, and finally, the outputs of the GAT and the CNN are spliced to obtain the final output of the CAC, namely the forwarding subgraph characteristics; the specific implementation is as shown in the formula:
wherein W is asA weight matrix is used on each node,
for the node characteristic value, weight is a one-dimensional convolution kernel, bias is represented by bias, and channel is represented by channel number;
furthermore, a new vector aggregation method Dynamic Routing-AT is provided in the step 2, and the research combines the Self-attribute weight calculation mode with the Dynamic Routing algorithm in the capsule network. The vector representation of the forwarding sub-graph is obtained after Dynamic Routing-AT aggregation, and preparation is also provided for the subsequent time information processing; the main formula of the Dynamic Routing-AT is:
wherein v is j For Dynamic Routing-AT one-time output, c ij Weight coefficient assigned to ith node, u i And (5) carrying out affine transformation on the vector of the user i.
Furthermore, in the step 3, the vector representation of the forwarding sub-graph obtained after the Dynamic Routing-AT in the step 2 is aggregated is sent into the GRU to process time information, the sub-graph is continuously divided along with the time, new nodes are continuously added into the diffusion process, and the time information is hidden in the sequence of the sub-graph;
still further, the system includes the following modules: the device comprises a data processing module, a feature extraction module, a feature aggregation module, a time information processing module and a prediction module. The system converts data into a forwarding graph form after the data is transmitted into a data processing module and transmits the data into a feature extraction module, the feature extraction module extracts forwarding graph features by using a cascade attention convolution network, then Dynamic Routing-AT aggregation forwarding sub-graph features are used in a feature aggregation module, a gating circulation unit is used in a time information processing module to process time information, and finally the time information is transmitted into a prediction module to apply a multi-layer sensor for final prediction. The system realizes the end-to-end public opinion hotspot information prediction;
compared with the prior art, the invention has the advantages that:
(1) The invention provides a brand new CAC-G system for predicting complex data of a social platform. The system takes the forwarding diagram as input and the final size of the forwarding diagram as output, is an end-to-end learning framework, and fully considers the structural information and time sequence information of the diagram.
(2) The invention creatively provides a model CAC for complex forwarding graph processing, and the model can pay more attention to global information by organically combining GAT and CNN, so that errors caused by information loss are reduced.
(3) The invention provides a new vector aggregation method Dynamic Routing-AT, and researches that a self-atttention weight calculation mode is combined with a Dynamic Routing algorithm in a capsule network, so that the stability of model prediction is stronger, and the time cost is greatly reduced.
Drawings
In order to describe the present invention in detail and clearly, examples and techniques used therein and related data will be described in the form of drawings, and it will be apparent to those skilled in the art that the present invention is not limited to the examples described in the present specification, but is not limited to the examples where the results are good.
FIG. 1 is a flowchart of a method and a system for predicting public opinion hotspot information according to an embodiment of the present disclosure;
FIG. 2 is an overall frame diagram followed by the operation of the embodiments of the present description;
FIG. 3 is a schematic diagram of a forwarding sub-graph through a GRU layer in an embodiment of the present disclosure;
Detailed Description
For a more detailed and clear explanation of the technology and advantages of the present invention, the embodiments and effects of the present invention will be described in further detail with reference to the accompanying drawings
A public opinion hotspot information prediction method and system, its overall structure block diagram is shown in figure 1; firstly, converting an observed message propagation process into a forwarding graph form, continuously dividing subgraphs according to the time passage into a model, considering the messy nature of social platform information in reality, providing a CAC model aiming AT complex forwarding graph processing, focusing on multi-scale information by CAC through the combination of GAT and CNN, transmitting node characteristics into CNN, transmitting edges and node characteristics into GAT together, performing splicing operation, then performing aggregation by using Dynamic Routing-AT to generate forwarding subgraph vector representation, transmitting the forwarding subgraph vector representation into time information hidden by GRU learning, and finally transmitting the time information into the MLP output cascade growth size. The method specifically comprises the following steps:
1. data collection
The invention uses the new wave microblog data set provided by the paper LRC-Q, the data set is grabbed from the new wave microblog and comprises the concerned relation among users and the forwarding links, when the user A forwards the information of the user B, the user A is called as the follower of the B, and the forwarding links from the B to the A exist, the data contains 30 ten popular information propagation processes, the data are eliminated by less than ten times of forwarding, the information is considered not propagated any more within twelve hours, three sub-data sets are constructed, the information propagation scale within 9 hours, 12 hours and 24 hours is predicted, and 70%,10% and 20% of each sub-data set are selected randomly on the basis to be used as training, verification and test data of a model.
2. (step 1) data preprocessing
Let g= (V, E) be a static social network, where V represents the set of users, E represents the set of edges, each message i propagated in the network forms an information cascade, subgraph C i ＝(V i ,E i ) Is defined as a cascade diagram, in which V i Representing a set of nodes through which a message passes, E i Representing connection V i At time t, if there is a message passing through the node, the state of the node is set to 1, otherwise 0, the node state is used
Representing and capturing a state diagram of a node in the cascade at time t>
Node state diagram to be captured->
And node status->
In combination, construct subgraph->
The structure of the cascade graph and the state of the node at the time t are contained.
3. (step 2) feature extraction of forwarding subgraphs by the Cascade attention convolutional network
The method has the advantages that the number of cascading graph edges is small or the repetition rate is high after the real social platform data are processed, so that a plurality of cascades are difficult to form a graph, the effect is not obvious when the graph neural network is processed, and then, CAC is provided for processing the cascading graph, wherein the CAC is formed by organically combining GAT and CNN, the design is that the GAT can sample all neighbor nodes, but the complexity of the social platform cascading network causes the neighbor nodes of a plurality of nodes to be undefined, the attention mechanism of the GAT is difficult to exert strong force, the phenomenon that the original information of the nodes is lost can be possibly caused, under the background, the CNN is introduced into a model, the CNN can pay attention to and extract the information of each node, the CNN is spliced with the node information extracted by the GAT, and the part of information possibly lost by the GAT is recovered. The working principle of CAC will be described in detail below.
The input to GAT is the edge and node characteristics, node characteristic matrix V ε R N×F Each row of (a) represents information of a node, and GAT calculates a node V by applying an autonomous semantic mechanism j For node V i Importance of (2)
As in equation 1; wherein N represents the number of nodes, F represents the characteristic dimension of the nodes, and W is the weight applied to each nodeMatrix (S)>
Is a node characteristic value. Then, using a LeakyReLU to perform nonlinearity, and finally normalizing neighbor nodes of the center node by softmax, wherein the neighbor nodes are shown in a graph-meaning network layer in FIG. 2, and the specific implementation is shown in a formula 2; in the present network, a multi-headed GAT is applied, and the formula under multi-headed attention is shown in formula 3.
The input of CNN is node characteristic, and the CNN is output out after passing through a layer of conv1d, and the specific formula is shown in formula 4; where weight is the one-dimensional convolution kernel, bias denotes offset, and channel denotes channel number. At the GAT layer output
And splicing the CNN layer output out to obtain CAC output B, wherein the specific formula is shown in formula 5.
Equation 1:
equation 2:
equation 3:
equation 4:
equation 5:
4. (step 3) Dynamic Routing-AT polymerization subgraph
The CAC outputs the characteristics of each node, and in order to better utilize the node characteristics and the hidden time characteristics, we aggregate the nodes in each forwarding sub-graph as the characteristic representation of the cascade graph. After being inspired by a Self-attribute weight calculation mode, the Self-attribute weight calculation mode is combined with a Dynamic Routing algorithm to provide a Dynamic Routing-AT, and the specific working principle is as follows:
vector u after affine transformation of user i i Output v with dynamic routing j And (5) performing dot multiplication to obtain a weight coefficient of the user i. Finally, a better effect can be achieved by only one iteration, which greatly reduces the time cost. The input of dynamic routing is the output node feature matrix B from CAC, we first perform linear affine transformation on the node representation vector, W is the mapping matrix, as shown in equation 1; when the node represents higher influence in the forwarding cascade subgraph, the weight coefficient c assigned to the i-th node ij The higher, where j represents the jth iteration, c ij The specific calculation mode of (2) is shown in a formula 2; dynamically routed output v j The specific formula of (2) is shown in formula 3.
Equation 1: u=wb
Equation 2:
equation 3:
5. (step 4) the gated loop unit learns the temporal feature
The effect of time effect in message popularity prediction is not negligible, for example, if a microblog message is widely browsed and forwarded at the first time just sent, the probability that the message becomes hot spot information is high. We can predict well the overall growth size of the cascade from the early forwarding time intervals, in the present invention, the time information is hidden in the sequence of forwarding cascade subgraphs, we will learn this part of the features using the GRU. When the model for predicting the time sequence is mainly based on a long-short-term memory network (LSTM) and GRU, the GRU is selected because compared with the LSTM, the GRU can achieve a comparable effect, and compared with the LSTM, the GRU is easier to train, so that the training efficiency can be greatly improved.
As shown in FIG. 3, the GRU is composed of an update gate and a geneticThe door is formed by forgetting to open, and specifically shown as a formula 1 and a formula 2; wherein r is t To reset the gate, z t To update the gate, σ is a sigmoid function, h t-1 For the hidden layer state before the moment t, W and U are weight matrixes, and the hidden state calculated by the gate is reset
As in equation 3; wherein->
Representing multiplication of corresponding elements in the operation matrix, the GRU obtains an output y by selectively forgetting original hidden states and selectively memorizing and summing information containing current nodes t And hidden state h t As shown in equation 4.
Equation 1: r is (r) t ＝σ(W r V t +U r h t-1 +b r )
Equation 2: z t ＝σ(W z V t +U z h t-1 +b z )
Equation 3:
equation 4:
6. (step 5) predicting growth Scale
The last module of the system is a prediction module, which outputs the output y of GRU t As an input to the MLP, the output of the MLP is of cascade growth scale, as in equation 1;
equation 1:
7. dividing the data set, constructing a loss function, training and testing the model
As shown in fig. 2, following the overall procedure, 70% of the microblog forwarding dataset was used as a training set, 20% was used as a validation set, and 10% was used as a test set. In this embodiment, mean square log transformation error (MSLE) was chosen as our experimentMeanwhile, MSLE is taken as an evaluation index of a model, and is specifically expressed as formula 1; where N represents the total number of cascades,
representing the true growth scale of the cascade, +.>
Representing the predicted cascade growth scale, T d Representing the time of observation.
Equation 1:
finally, the system predicts the hot spot information, compared with the prior model, the method can extract useful information more accurately when facing the data of the noisy social platform, so that the prediction accuracy is greatly improved, meanwhile, the training time is greatly reduced, the efficiency of hot spot information prediction is greatly improved, and a brand new thought is provided for the popularity prediction field.
In summary, although embodiments of the present invention have been described, these are merely preferred embodiments of the present invention, and are not intended to be limiting in any way, and it will be apparent to those skilled in the art that various modifications and variations and combinations of technical contents of the present invention can be made to achieve equivalent results, and these means still fall within the scope of the present invention.
Claims (5)
1. A public opinion hotspot information prediction method and system are characterized by comprising the following steps:
step 1: the method comprises the steps of continuously dividing a converted graph structure into forwarding subgraphs according to the time lapse, learning subgraph features through a CAC model, wherein the CAC model organically combines GAT and CNN, sends node features and edge features into GAT for learning, inputs the node features into CNN, and finally performs splicing operation on the outputs of the GAT and CNN to obtain output features of the forwarding subgraphs;
step 2: combining a Self-attribute weight calculation mode with a Dynamic Routing algorithm in a capsule network to provide Dynamic Routing-AT, and sending forwarding sub-graph features output by CAC into the Dynamic Routing-AT for aggregation to obtain vector representation of the sub-graph;
step 3: the invention uses GRU to process time information, the subgraph is divided continuously along with the time, new nodes are added into the diffusion process continuously, the time information is hidden in the subgraph sequence processed by Dynamic Routing-AT, and the prediction is carried out after the time information is processed;
step 4: predicting the forwarding cascade data scale through the constructed model;
and testing the model, and putting part of data on the model to test by using the parameters learned during training to obtain a test result.
2. The public opinion hotspot information prediction method and system according to claim 1, wherein in the step 1, a model CAC for complex data processing of a social platform is innovatively provided, and GAT and CNN are organically combined. The GAT with multiple head attentions is used in the CAC, the node characteristics and the edge characteristics of the forwarding subgraph are input into the GAT, meanwhile, the node characteristics are input into the CNN, and finally, the outputs of the GAT and the CNN are spliced to obtain the final output of the CAC, namely the forwarding subgraph characteristics; the specific implementation is as shown in the formula:
where W is the weight matrix applied to each node,
for node eigenvalues, weight is a one-dimensional convolution kernel, bias represents bias, channel represents channel number.
3. The public opinion hotspot information prediction method and system according to claim 1, wherein a new vector aggregation method Dynamic Routing-AT is provided in the step 2, and the research combines a Self-attribute weight calculation mode with a Dynamic Routing algorithm in a capsule network. The vector representation of the forwarding sub-graph is obtained after Dynamic Routing-AT aggregation, and preparation is also provided for the subsequent time information processing; the main formula of the Dynamic Routing-AT is:
wherein v is j For Dynamic Routing-AT one-time output, c ij Weight coefficient assigned to ith node, u i And (5) carrying out affine transformation on the vector of the user i.
4. The public opinion hotspot information prediction method and system according to claim 1, wherein in the step 3, vector representations of forwarding subgraphs obtained by aggregating Dynamic Routing-ATs in the step 2 are sent to a GRU to process time information, subgraphs are divided continuously along with the time, new nodes are added to a diffusion process continuously, and the time information is hidden in the sequence of subgraphs.
5. The public opinion hotspot information prediction method and system as claimed in claim 1, comprising the following modules: the device comprises a data processing module, a feature extraction module, a feature aggregation module, a time information processing module and a prediction module. The system converts data into a forwarding graph form after the data is transmitted into a data processing module and transmits the data into a feature extraction module, the feature extraction module extracts forwarding graph features by using a cascade attention convolution network, then Dynamic Routing-AT aggregation forwarding sub-graph features are used in a feature aggregation module, a gating circulation unit is used in a time information processing module to process time information, and finally the time information is transmitted into a prediction module to apply a multi-layer sensor for final prediction. The system realizes the end-to-end public opinion hotspot information prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211244827.8A CN116307022A (en) | 2022-10-12 | 2022-10-12 | Public opinion hotspot information prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211244827.8A CN116307022A (en) | 2022-10-12 | 2022-10-12 | Public opinion hotspot information prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116307022A true CN116307022A (en) | 2023-06-23 |
Family
ID=86822745
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211244827.8A Pending CN116307022A (en) | 2022-10-12 | 2022-10-12 | Public opinion hotspot information prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116307022A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117610717A (en) * | 2023-11-13 | 2024-02-27 | 重庆大学 | Information popularity prediction method based on double-variation cascade self-encoder |
-
2022
- 2022-10-12 CN CN202211244827.8A patent/CN116307022A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117610717A (en) * | 2023-11-13 | 2024-02-27 | 重庆大学 | Information popularity prediction method based on double-variation cascade self-encoder |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110070183B (en) | Neural network model training method and device for weakly labeled data | |
| CN110851782A (en) | Network flow prediction method based on lightweight spatiotemporal deep learning model | |
| CN114818515A (en) | Multidimensional time sequence prediction method based on self-attention mechanism and graph convolution network | |
| CN111126674A (en) | Propagation prediction method and system based on depth model | |
| CN115512545B (en) | Traffic speed prediction method based on space-time dynamic graph convolution network | |
| CN114519469A (en) | Construction method of multivariate long sequence time sequence prediction model based on Transformer framework | |
| Dai et al. | Hybrid deep model for human behavior understanding on industrial internet of video things | |
| Dai et al. | Spatio-temporal deep learning framework for traffic speed forecasting in IoT | |
| CN115659966A (en) | Rumor detection method and system based on dynamic heteromorphic graph and multi-level attention | |
| Xu et al. | Short‐term traffic flow prediction based on whale optimization algorithm optimized BiLSTM_Attention | |
| CN116307022A (en) | Public opinion hotspot information prediction method and system | |
| CN117272195A (en) | Block chain abnormal node detection method and system based on graph convolution attention network | |
| CN114821218A (en) | Target detection model searching method based on improved channel attention mechanism | |
| CN117236492B (en) | Traffic demand prediction method based on dynamic multi-scale graph learning | |
| CN116993185A (en) | Time sequence prediction method, device, equipment and storage medium | |
| CN116258504B (en) | Bank customer relationship management system and method thereof | |
| CN114357160B (en) | Early rumor detection method and device based on generated propagation structural features | |
| Zhang et al. | Obstacle‐transformer: A trajectory prediction network based on surrounding trajectories | |
| WO2023137918A1 (en) | Text data analysis method and apparatus, model training method, and computer device | |
| CN111935259B (en) | Method and device for determining target account set, storage medium and electronic equipment | |
| CN114139805A (en) | Temperature prediction and model training method and related device | |
| Cai et al. | Efficient Split Learning with Non-iid Data | |
| Bi et al. | Multi-indicator water time series imputation with autoregressive generative adversarial networks | |
| Zhang et al. | Image caption generation method based on an interaction mechanism and scene concept selection module | |
| Zhang et al. | Skeleton‐Guided Action Recognition with Multistream 3D Convolutional Neural Network for Elderly‐Care Robot |
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 |