CN113946680A - Online network rumor identification method based on graph embedding and information flow analysis - Google Patents
Online network rumor identification method based on graph embedding and information flow analysis Download PDFInfo
- Publication number
- CN113946680A CN113946680A CN202111219029.5A CN202111219029A CN113946680A CN 113946680 A CN113946680 A CN 113946680A CN 202111219029 A CN202111219029 A CN 202111219029A CN 113946680 A CN113946680 A CN 113946680A
- Authority
- CN
- China
- Prior art keywords
- network
- information
- graph
- information flow
- public
- 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 40
- 238000005206 flow analysis Methods 0.000 title claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 238000013507 mapping Methods 0.000 claims abstract description 13
- 230000007246 mechanism Effects 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 239000010410 layer Substances 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000000306 recurrent effect Effects 0.000 claims description 4
- 239000002356 single layer Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000009193 crawling Effects 0.000 claims description 2
- 230000002457 bidirectional effect Effects 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 11
- 230000001364 causal effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 244000097202 Rathbunia alamosensis Species 0.000 description 1
- 235000009776 Rathbunia alamosensis Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an online network rumor identification method based on graph embedding and information flow analysis, which has the technical scheme that: firstly, deeply preprocessing an online network public sentiment event, shaving interference data, retaining characteristic information, and mapping the characteristic information to a high-dimensional space with word meaning relevance; then, based on a graph embedding method, extracting public opinion network structure characteristics contained in the hierarchical relation graph by using a deep graph convolution neural network; then establishing a bidirectional gate cycle neural network with a depth structure, and analyzing bidirectional time sequence characteristics in public sentiment information flow; and finally, fusing the network structure characteristics and the bidirectional time sequence characteristics of the public sentiments by using an attention mechanism, thereby improving the accuracy of identifying the false information in the online network public sentiment information.
Description
Technical Field
The invention belongs to the technical field of deep learning and rumor identification, and particularly relates to an online network rumor identification method based on graph embedding and information flow analysis.
Background
The online social network provides great convenience for information diffusion, and simultaneously builds a virtual bridge for rumor accelerated propagation. The spread of rumors causes huge economic losses on one hand and threatens harmonious and stable operation of the society on the other hand, and becomes a social phenomenon which has to be given enough attention. The rumors generally refer to information which is not based on facts or intentionally forged, and the traditional rumors identification adopts a manual method, but nowadays of information explosion, the attempt to identify numerous online public opinions by manpower is almost an impossible task. In recent years, with the push of information technology development, academics and industry have begun to shift the view of rumor identification to an automated machine learning method. A very intuitive identification method is to analyze the public opinion information, and the method only focuses on the syntax and semantic content carried by the public opinion information and extracts high-dimensional linguistic features by means of a modern machine learning algorithm. However, due to the limitation of social platforms such as microblog and Twitter on the length of text, such feature-based methods are difficult to extract enough features, thereby limiting the accuracy of rumor identification. In order to overcome the difficulty in feature extraction, some scholars begin to add subsequent information such as comments and forwarding of original public opinion information to the task of public opinion identification. In the method, a tree structure for information propagation is established classically, and then a backtracking method is adopted to backtrack from leaf nodes to a root node of an information source step by step. However, such algorithms are not very competitive for early rumor identification, limited by the imperfection of the propagation tree at the early stage of public opinion propagation.
Disclosure of Invention
In order to overcome the defects of the method, the invention provides an online network rumor identification method based on graph embedding and information flow analysis, which adds graph embedding for a propagation space structure while considering public sentiment information time sequence characteristics, and then fuses the graph embedding and the propagation space structure by utilizing an attention mechanism to form a dynamically weighted high-dimensional public sentiment characteristic representation system which is characterized by data driving, thereby improving the identification accuracy of the online network rumor.
In order to achieve the above purpose, the present invention achieves the identification of online network rumors by the following technical solutions:
the present invention is defined as follows:
definition 1 public opinion information sources
A public opinion information source is an interpretation of original information about a particular event published by individuals on a network. Formally, a source of public sentiment information appears at the earliest on an information distribution page of an individual or group, using sjRepresenting a source of public sentiment information, and the subscript j refers to a particular public sentiment event.
Definition 2 reaction information
The reaction information is the reaction of people to the public sentiment information source, and is inevitably later than the public sentiment information source in the release time, and is used
The response information is shown, and the superscript n and the subscript j respectively refer to the sequence of the response information on the release time and a specific public sentiment event.
Defining 3 public sentiment information flows
The public opinion information flow is an information set formed by all public opinion information in the time from a public opinion information source to the last reaction information release. In a determined information flow
In (| P | represents the number of information), the information is sorted according to the issue time. Thus, all information streams form the whole public opinion suspected set T ═ T { (T) about a plurality of events1,T2,…,TNN represents all events.
Defining 4 public opinion hierarchical relation diagram
The public opinion hierarchical relation graph is a hierarchical network formed on a topological space according to the causal relation of public opinion information distribution. Using VjAnd EjRepresenting edges and points in the network, then the public opinion hierarchical relationship graph with self-connecting rings can be represented as Gj＝{Vj,Ej；φj,ψjIn which phijRepresenting causal relationships between public opinion information, psijRepresenting a self-connecting loop. Thus, all public opinion relationship graphs can be expressed as G ═ G1,G2,…,GNN represents all events.
An online network rumor identification method based on graph embedding and information flow analysis is characterized in that: firstly, deeply preprocessing an online network public sentiment event, shaving interference data, retaining characteristic information, and mapping the characteristic information to a high-dimensional space with word meaning relevance; then, based on a graph embedding method, extracting public opinion network structure characteristics contained in the hierarchical relation graph by using a deep graph convolution neural network; then establishing a bidirectional gate cycle neural network with a depth structure, and analyzing bidirectional time sequence characteristics in public sentiment information flow; and finally, fusing the network structure characteristics and the bidirectional time sequence characteristics of the public sentiments by using an attention mechanism, thereby improving the accuracy of identifying the false information in the online network public sentiment information.
The invention is realized by the following specific scheme:
an online network rumor identification method based on graph embedding and information flow analysis mainly comprises the following steps:
step 1) determining an online network public opinion data source, crawling structured public opinion data, and preprocessing the obtained public opinion data;
step1, deleting invalid information such as blank tweets, meaningless comments, topic labels and the like;
step2, mapping all the emoticons in the public sentiment information into a textual expression according to the emoticon interpretations of the network standard;
step3, carrying out word segmentation processing on the public sentiment information by utilizing Jieba and space, and deleting stop words at the same time;
step4, based on the SOTA algorithm, adopting Word Embedding method to map the Word information to the high-dimensional space with spatial relevance.
Step 2) analyzing the spatial structural characteristics of online network public opinion data, abstracting an issuing source of original information into a central node of a network, abstracting subsequent comment, forwarding and reply information into relational nodes on the network, establishing a hierarchical structure among the nodes in the network based on the logical relationship of information issuing, mapping the constructed hierarchical network into an undirected graph structure with self-connection rings, and extracting the structural characteristics of a public opinion propagation network by a graph embedding method by means of a graph convolution neural network GCN;
step1, mapping the constructed hierarchical network into a network with self-connecting rings INUndirected graph G ofjThe adjacency matrix of the original hierarchical network is represented by A%, and an undirected graph G with self-connected loops is calculatedjThe adjacent matrix a of (2) is calculated by:
A＝A％+IN
step2 calculation of undirected graph GjThe degree distribution matrix D of (2) is calculated by the following method:
D＝∑kAk
step3, carrying out graph convolution operation by utilizing the idea of Fourier transform in the graph convolution neural network GCN, and carrying out first-order approximation on the obtained convolution graph to obtain an approximation graph convolution value L of a single-layer network(1)The calculation method comprises the following steps:
wherein X is Word Embedding value of public sentiment information, W(1)Is a weight matrix in the first layer of the GCN network,
δ (g) is the activation function, usually the ReLU function is chosen;
step4, limiting the limitation of single-layer graph convolution neural network GCN to the network space structure, adding a layer of graph convolution layer, performing convolution operation, and updating the convolution value L twice(2)Namely, the extracted public opinion propagation network structural features are calculated as follows:
wherein W(2)Is a self-learning weight matrix in the second layer of the GCN network.
Step 3) extracting the release time of each information in the online network public opinion data, constructing a public opinion information flow by taking the release time as a basis, establishing a Bi-directional gate circulation neural network Bi-GRU with a deep structure by taking the obtained information flow as input, respectively extracting the time sequence characteristics in the public opinion information in the forward and reverse time sequence angles, and fusing the time sequence characteristics in two dimensions to form the angle overall characteristic presentation of the public opinion information flow in the time sequence;
step1, the public sentiment events are sorted according to the release time, and a time sequence information flow with a more complete description for a certain public sentiment event is established.
Step2, inputting public sentiment information into the recurrent neural network GRU according to the forward information flow direction, and extracting the time sequence characteristic h 'contained in the forward information flow'a,tCalculated by the following formula:
wherein, Uz,Wz,Ur,Wr,Uh％And Wh％Respectively represent learnable weights within a GRU unit, e represents element-by-element multiplication operation, h'a,t-1Is a time sequence characteristic of the last point in time,
is h'a,tThe hidden layer candidate state of (a) is,
represents the states of the reset gate and the update gate, respectively, σ (·) being a Sigmoid function;
step3, inputting public sentiment information into the gated recurrent neural network GRU in sequence according to the reverse information flow direction, wherein the extracted reverse information flow contains the time sequence characteristic of h'a,tThe calculation method is the same as that in the forward direction;
step4 connectingCombining the forward and reverse information flow time sequence characteristics to form the overall characteristic presentation h of the public sentiment information flow on the time sequencea,t。
Step 4) adopting an Attention mechanism, fusing graph structure characteristics obtained by graph embedding and overall presented information flow time sequence characteristics to form a dynamically weighted high-dimensional public opinion characteristic representation system which is characterized by data driving, and taking the output of the representation system as a final mixed characteristic presentation mode for rumor identification;
step1, establishing a gate cycle neural network with the same parameters as those in the Step 3), and carrying out structural characteristics L of the public sentiment propagation network based on the graph embedding method obtained in the Step 2)(2)As input, a time series characteristic h of a consensus information stream is obtaineda,tNetwork output h with identical dimensionsb,t；
Step2 fusion of h by Attention mechanisma,tAnd hb,tTo obtain htNamely the final mixed feature presentation, the specific calculation method is as follows:
wherein, γ and WhIs a learnable weight parameter in the network.
And step 5) taking the fully-connected network as an output layer, mapping output to a probability space by matching with a Softmax function, defining a Loss function, judging a virtual space distance between a probability result predicted by the model and real data, taking the minimization of the Loss function as an optimization direction of the model, and continuously adjusting each related weight parameter in the network.
The Loss function is:
wherein N issIs the sample size, C is the sample classification, yc(xn) Is the number of samples classified into a certain category,
is the number of samples predicted by the model as a specific category, | | θ | | | Y2Is the second order norm of all learnable parameters to avoid over-fitting.
Drawings
Fig. 1 is a flow chart illustrating an online network rumor identification method based on graph embedding and information flow analysis.
Detailed Description
In order to make the technical route, the structural features and the achievement effect of the invention clear, the invention is further described in the following with the accompanying drawings and specific examples so that the related art can better understand and implement the invention, but the used examples are not limiting on the invention.
As shown in fig. 1, an online network rumor identification method based on graph embedding and information flow analysis mainly includes the following steps:
step 1) selecting two typical online social networking platforms of a Xinlang microblog (Chinese) platform and a Twitter platform (English) platform, respectively acquiring public related network published data about a series of public sentiment events, and deleting invalid information such as blank tweets, meaningless comments, topic labels and the like in the data; then mapping all the emoticons in the public sentiment information into a written expression according to the emoticon interpretations of the network standard; carrying out word segmentation processing on the public sentiment information by utilizing Jieba and space, and deleting stop words at the same time; finally, on the basis of the SOTA algorithm, Word Embedding is adopted to map Word information to a high-dimensional space with spatial relevance, and the Word Embedding dimension is set to be 300;
step 2) analyzing the spatial structural characteristics of the acquired public opinion data on the Xinlang microblog and the Twitter respectively, abstracting the publishing source of the original information as a central node of the network, abstracting the subsequent comment, forwarding and reply information as a relational node on the network, establishing a hierarchical structure among nodes in the network based on the logical relationship of information publishing, and mapping the constructed hierarchical network into an undirected graph structure with a self-connection ring;
constructing an adjacency matrix according to the established self-loop undirected graph mechanism, wherein if the causal relationship of information release exists between two adjacent nodes in the graph, the value in the adjacency matrix is marked as 1, otherwise, the value is 0; because the adjacent matrix is of a sparse structure, a construction mode of the sparse matrix is used, so that the memory consumption is saved;
establishing a degree distribution matrix according to the adjacent matrix;
constructing a graph convolution neural network to extract structural features of a public sentiment propagation network, wherein a graph embedding mode is carried out by using a two-layer stacked GCN structure, the related Fourier transform adopts first-order approximation, the hidden layer dimension in the GCN is set to be 256, and the output layer dimension is set to be 128;
and 3) sequencing the public sentiment events on the Sina microblog and the Twitter according to the release time of the public sentiment events, and establishing a time sequence information flow which has relatively complete description on a certain public sentiment event. Then the obtained information flow is used as input, a Bi-directional gate circulation neural network Bi-GRU with a depth structure is established, time sequence characteristics in public sentiment information in forward and reverse time sequence angles are respectively extracted, and then the time sequence characteristics in two dimensions are fused to form overall characteristic presentation of the public sentiment information flow in the time sequence angle;
in the implementation process of the gate cycle neural network GRU, Dropout operation is required to be added to prevent overfitting; considering the inconsistency of the public opinion information length, Padding operation is required to be carried out on the information flow, so that the information in the specific public opinion event is unified into the same length, a Padding value is set to be 0, the number of Bi-GRU layers is set to be 2, and the hidden layer dimension is 256;
step 4) establishing a gate cycle neural network with the same parameters as those in the step 3), taking the structural characteristics of the public sentiment propagation network based on the graph embedding method obtained in the step 2) as input, and carrying out dimension transformation on the input to obtain a network output with the same dimension as the time sequence characteristic of the public sentiment information flow;
an Attention attribution mechanism is adopted, graph structure characteristics obtained by graph embedding and overall presented information flow time sequence characteristics are fused, a dynamically weighted high-dimensional public opinion characteristic representation system which is characterized by data driving is formed, and the output of the representation system is used as a rumor to identify a final mixed characteristic presentation mode; the activation function of the Attention network adopts Tanh, the dimension of an input layer is 256 x 2, the output is Attention Score, and the dimension is 1;
and step 5) taking the fully-connected network as an output layer, mapping output to a probability space by matching with a Softmax function, defining a Loss function, judging a virtual space distance between a probability result predicted by the model and real data, taking the minimization of the Loss function as an optimization direction of the model, and continuously adjusting each related weight parameter in the network.
The model's Batch Size is set to 128 and the optimizer chooses Adam, β1And beta2Set to 0.9 and 0.999, respectively; using a random gradient descending method, setting the initial learning rate to be 2e-3, and gradually descending along with the training process;
the evaluation indexes adopt Accuracy, Precision, Recall and F1 score, and the specific calculation formula is as follows:
wherein TP is the number of positive cases with correct prediction, FP is the number of positive cases with wrong prediction, TN is the number of negative cases with correct prediction, and FN is the number of negative cases with wrong prediction;
for predictive analysis of rumors in public opinion, 4-classification subdivision was used, non-rumors (NR), pseudo-rumors (FR), True Rumors (TR) and Unverified Rumors (UR);
the results of comparative experiments with the identification of related rumors are shown in Table 1 below
TABLE 1 rumor identification comparison experiment results
Microblog data set
Twitter data set
It can be seen that the effect of the invention on rumor identification is the highest under each category, and compared with the comparison method, the accuracy of rumor identification in online public opinion events is improved.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (6)
1. An online network rumor identification method based on graph embedding and information flow analysis is characterized by comprising the following specific steps:
step 1) determining an online network public opinion data source, crawling structured public opinion data, and preprocessing the obtained public opinion data;
step 2) analyzing the spatial structural characteristics of online network public opinion data, abstracting an issuing source of original information into a central node of a network, abstracting subsequent comment, forwarding and reply information into relational nodes on the network, establishing a hierarchical structure among the nodes in the network based on the logical relationship of information issuing, mapping the constructed hierarchical network into an undirected graph structure with self-connection rings, and extracting the structural characteristics of a public opinion propagation network by a graph embedding method by means of a graph convolution neural network GCN;
step 3) extracting the release time of each information in the online network public opinion data, constructing a public opinion information flow by taking the release time as a basis, establishing a Bi-directional gate circulation neural network Bi-GRU with a deep structure by taking the obtained information flow as input, respectively extracting the time sequence characteristics in the public opinion information in the forward and reverse time sequence angles, and fusing the time sequence characteristics in the two dimensions to form the overall characteristic presentation of the public opinion information flow in the time sequence angle;
step 4) adopting an Attention mechanism, fusing graph structure characteristics obtained by graph embedding and overall presented information flow time sequence characteristics to form a dynamically weighted high-dimensional public opinion characteristic representation system which is characterized by data driving and taking the system as a rumor to identify a final mixed characteristic presentation mode;
and step 5) taking the fully-connected network as an output layer, mapping output to a probability space by matching with a Softmax function, defining a Loss function, judging a virtual space distance between a probability result predicted by the model and real data, taking the minimization of the Loss function as an optimization direction of the model, and continuously adjusting each related weight parameter in the network.
2. The method of claim 1, wherein the step 1) of pre-processing online public opinion data comprises the following steps:
step1, deleting invalid information such as blank tweets, meaningless comments, topic labels and the like;
step2, mapping all the emoticons in the public sentiment information into a textual expression according to the emoticon interpretations of the network standard;
step3, carrying out word segmentation processing on the public sentiment information by utilizing Jieba and space, and deleting stop words at the same time;
step4, based on the SOTA algorithm, adopting Word Embedding method to map the Word information to the high-dimensional space with spatial relevance.
3. The method of claim 1, wherein the step 2) of extracting the structural features of the public opinion propagation network based on the graph embedding method comprises:
step1, mapping the constructed hierarchical network into a network with self-connecting rings INUndirected graph G ofjThe adjacency matrix of the original hierarchical network is represented by A%, and an undirected graph G with self-connected loops is calculatedjThe adjacent matrix a of (2) is calculated by:
step2 calculation of undirected graph GjThe degree distribution matrix D of (2) is calculated by the following method:
D＝∑kAk
step3, carrying out graph convolution operation by utilizing the idea of Fourier transform in the graph convolution neural network GCN, and carrying out first-order approximation on the obtained convolution graph to obtain an approximation graph convolution value L of a single-layer network(1)The calculation method comprises the following steps:
wherein X is Word Embedding value of public sentiment information, W(1)Is a weight matrix in the first layer of the GCN network,
δ (g) is the activation function, usually the ReLU function is chosen;
step4, limiting the limitation of single-layer graph convolution neural network GCN to the network space structure, adding a layer of graph convolution layer, performing convolution operation, and updating the convolution value L twice(2)Namely, the extracted public opinion propagation network structural features are calculated as follows:
wherein W(2)Is a self-learning weight matrix in the second layer of the GCN network.
4. The method of claim 1, wherein the general characteristics of the public opinion information flow in time sequence are presented in step 3), and the method comprises:
step1, the public sentiment events are sorted according to the release time, and a time sequence information flow which has more complete description for a certain public sentiment event is established;
step2, inputting public sentiment information into the recurrent neural network GRU according to the forward information flow direction, and extracting the time sequence characteristic h 'contained in the forward information flow'a,tCalculated by the following formula:
wherein, Uz,Wz,Ur,Wr,
And
respectively represent learnable weights within a GRU unit, e represents element-by-element multiplication operation, h'a,t-1Is a time sequence characteristic of the last point in time,
is h'a,tThe hidden layer candidate state of (a) is,
representing the states of the reset gate and the update gate, respectively, σ (-) being Sigmoid a function;
step3, inputting public sentiment information into the gate recurrent neural network GRU in sequence according to the reverse information flow direction, wherein the extracted reverse information flow contains a time sequence characteristic h ″a,tThe calculation method is the same as that in the forward direction;
step4, connecting the forward and reverse information flow time sequence characteristics to form the overall characteristic presentation h of the public sentiment information flow on the time sequencea,t。
5. The method for identifying online network rumors based on graph embedding and information flow analysis of claim 1, wherein step 4) employs a mixed feature presentation mode of Attention mechanism fusion, and the specific method is as follows:
step1, establishing a gate cycle neural network with the same parameters as those in the Step 3), and carrying out structural characteristics L of the public sentiment propagation network based on the graph embedding method obtained in the Step 2)(2)As input, a time series characteristic h of a consensus information stream is obtaineda,tNetwork output h with identical dimensionsb,t；
Step2 fusion of h by Attention mechanisma,tAnd hb,tTo obtain htNamely the final mixed feature presentation, the specific calculation method is as follows:
wherein, γ and WhIs a learnable weight parameter in the network.
6. The method for identifying online network rumors based on graph embedding and information flow analysis of claim 1, wherein the step 5) of determining the Loss function is specifically as follows:
wherein N issIs the sample size, C is the sample classification, yc(xn) Is the number of samples classified into a certain category,
is the number of samples predicted by the model as a specific category, | | θ | | | Y2Is the second order norm of all learnable parameters to avoid over-fitting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111219029.5A CN113946680B (en) | 2021-10-20 | 2021-10-20 | Online network rumor identification method based on graph embedding and information flow analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111219029.5A CN113946680B (en) | 2021-10-20 | 2021-10-20 | Online network rumor identification method based on graph embedding and information flow analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113946680A true CN113946680A (en) | 2022-01-18 |
| CN113946680B CN113946680B (en) | 2024-04-16 |
Family
ID=79331674
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111219029.5A Active CN113946680B (en) | 2021-10-20 | 2021-10-20 | Online network rumor identification method based on graph embedding and information flow analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113946680B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685153A (en) * | 2018-12-29 | 2019-04-26 | 武汉大学 | A kind of social networks rumour discrimination method based on characteristic aggregation |
| WO2019205318A1 (en) * | 2018-04-25 | 2019-10-31 | 平安科技（深圳）有限公司 | Public opinion information classification method and apparatus, computer device, and storage medium |
| WO2019227710A1 (en) * | 2018-05-31 | 2019-12-05 | 平安科技（深圳）有限公司 | Network public opinion analysis method and apparatus, and computer-readable storage medium |
| WO2020000847A1 (en) * | 2018-06-25 | 2020-01-02 | 中译语通科技股份有限公司 | News big data-based method and system for monitoring and analyzing risk perception index |
| CN112069397A (en) * | 2020-08-21 | 2020-12-11 | 三峡大学 | Rumor detection method combining self-attention mechanism with generation of confrontation network |
| CN112231562A (en) * | 2020-10-15 | 2021-01-15 | 北京工商大学 | Network rumor identification method and system |
| WO2021073271A1 (en) * | 2019-10-17 | 2021-04-22 | 平安科技（深圳）有限公司 | Public opinion analysis method and device, computer device and storage medium |
| CN112766359A (en) * | 2021-01-14 | 2021-05-07 | 北京工商大学 | Word double-dimensional microblog rumor recognition method for food safety public sentiment |
| CN113204641A (en) * | 2021-04-12 | 2021-08-03 | 武汉大学 | Annealing attention rumor identification method and device based on user characteristics |
-
2021
- 2021-10-20 CN CN202111219029.5A patent/CN113946680B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019205318A1 (en) * | 2018-04-25 | 2019-10-31 | 平安科技（深圳）有限公司 | Public opinion information classification method and apparatus, computer device, and storage medium |
| WO2019227710A1 (en) * | 2018-05-31 | 2019-12-05 | 平安科技（深圳）有限公司 | Network public opinion analysis method and apparatus, and computer-readable storage medium |
| WO2020000847A1 (en) * | 2018-06-25 | 2020-01-02 | 中译语通科技股份有限公司 | News big data-based method and system for monitoring and analyzing risk perception index |
| CN109685153A (en) * | 2018-12-29 | 2019-04-26 | 武汉大学 | A kind of social networks rumour discrimination method based on characteristic aggregation |
| WO2021073271A1 (en) * | 2019-10-17 | 2021-04-22 | 平安科技（深圳）有限公司 | Public opinion analysis method and device, computer device and storage medium |
| CN112069397A (en) * | 2020-08-21 | 2020-12-11 | 三峡大学 | Rumor detection method combining self-attention mechanism with generation of confrontation network |
| CN112231562A (en) * | 2020-10-15 | 2021-01-15 | 北京工商大学 | Network rumor identification method and system |
| CN112766359A (en) * | 2021-01-14 | 2021-05-07 | 北京工商大学 | Word double-dimensional microblog rumor recognition method for food safety public sentiment |
| CN113204641A (en) * | 2021-04-12 | 2021-08-03 | 武汉大学 | Annealing attention rumor identification method and device based on user characteristics |
Non-Patent Citations (2)
| Title |
|---|
| 王晰巍;张柳;韦雅楠;王铎;: "社交网络舆情中意见领袖主题图谱构建及关系路径研究――基于网络谣言话题的分析", 情报资料工作, no. 02, 25 March 2020 (2020-03-25) * |
| 王淼;郭阳明;陈泽林;钟林龙;: "基于RNN的网络舆情谣言判断研究", 电脑知识与技术, no. 24, 25 August 2020 (2020-08-25) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113946680B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Priyadarshini et al. | A novel LSTM–CNN–grid search-based deep neural network for sentiment analysis | |
| CN108875051B (en) | Automatic knowledge graph construction method and system for massive unstructured texts | |
| Vateekul et al. | A study of sentiment analysis using deep learning techniques on Thai Twitter data | |
| CN109726745B (en) | Target-based emotion classification method integrating description knowledge | |
| Hassan et al. | Sentiment analysis on bangla and romanized bangla text (BRBT) using deep recurrent models | |
| CN110851593B (en) | Complex value word vector construction method based on position and semantics | |
| Azim et al. | Text to emotion extraction using supervised machine learning techniques | |
| CN113407660A (en) | Unstructured text event extraction method | |
| CN115017887A (en) | Chinese rumor detection method based on graph convolution | |
| CN113094502A (en) | Multi-granularity takeaway user comment sentiment analysis method | |
| Al Maruf et al. | Emotion detection from text and sentiment analysis of Ukraine Russia war using machine learning technique | |
| CN116010553A (en) | Viewpoint retrieval system based on two-way coding and accurate matching signals | |
| Jayakody et al. | Sentiment analysis on product reviews on twitter using Machine Learning Approaches | |
| Soni et al. | The use of supervised text classification techniques: A comprehensive study | |
| Başarslan et al. | Sentiment analysis with ensemble and machine learning methods in multi-domain datasets | |
| CN116245110A (en) | Multi-dimensional information fusion user standing detection method based on graph attention network | |
| Shah et al. | A comprehensive study and detailed review on hate speech classification: a systematic analysis | |
| CN116227486A (en) | Emotion analysis method based on retrieval and contrast learning | |
| CN113946680B (en) | Online network rumor identification method based on graph embedding and information flow analysis | |
| Jasim et al. | Analyzing Social Media Sentiment: Twitter as a Case Study | |
| Fadel et al. | A comparative study for supervised learning algorithms to analyze sentiment tweets | |
| Meenakshi et al. | Deep learning techniques for spamming and cyberbullying detection | |
| Wang et al. | BiLSTM-ATT Chinese sentiment classification model based on pre-training word vectors | |
| CN113704409B (en) | False recruitment information detection method based on cascading forests | |
| Purwitasari et al. | Word-Embedding Model for Evaluating Text Generation of Imbalanced Spam Reviews |
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 |