CN110210016A - Bilinearity neural network Deceptive news detection method and system based on style guidance - Google Patents
Bilinearity neural network Deceptive news detection method and system based on style guidance Download PDFInfo
- Publication number
- CN110210016A CN110210016A CN201910341056.6A CN201910341056A CN110210016A CN 110210016 A CN110210016 A CN 110210016A CN 201910341056 A CN201910341056 A CN 201910341056A CN 110210016 A CN110210016 A CN 110210016A
- Authority
- CN
- China
- Prior art keywords
- vector
- style
- text
- neural network
- bilinearity
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Machine Translation (AREA)
Abstract
The present invention proposes a kind of bilinearity neural network Deceptive news detection method and system based on style guidance, it include: the newsletter archive for obtaining and being detected to network false news, quantify the diction feature of the newsletter archive by neural network, obtain the style vector of the newsletter archive, the newsletter archive is inputted into Text character extraction device, obtains the text vector of the newsletter archive;The style vector sum text vector is inputted into bilinearity neural network, the bilinearity neural network includes bilinear function, for modeling the correlation between the style vector sum text vector, to obtain style-text feature matrix of the newsletter archive, boot vector is formed using largest score vector in the style-text feature matrix, and the boot vector is input to full articulamentum, determine the Deceptive news label of the newsletter archive.The present invention guides the learning process of deep learning model according to the diction of Deceptive news general character, improves recognition accuracy and the Generalization Capability of model.
Description
Technical field
The present invention relates to big data excavate in news detection field, and it is in particular to a kind of based on style guidance bilinearity
Neural network Deceptive news detection method and system.
Background technique
The fast development of social media has changed daily life, and user can be convenient freely from social matchmaker
Publication and acquisition information on body.However, the booming of social media also provides fertile soil for the growth of Deceptive news and propagation.
According to statistics, only just there are 529 Deceptive news in relation to presidential candidate to generate during US presidential election in 2016, be transmitted
Up to 3,007,000,000 times.Deceptive news seriously polluted network social intercourse environment, affect the daily life of user, therefore
It needs to carry out automatic detection to the Deceptive news on network social intercourse media.
In existing research, researcher is generally concerned on news content and corresponding social networks.Martin et al. benefit
Deceptive news detection is carried out with the domain dependant informations such as URL number in reference word number, news, Jin et al. passes through inspection
Whether there is conflict between the different viewpoints surveyed under discovery news related commentary to determine the authenticity of news.Castillo et al. is from new
It hears content, user home page and dissemination of news network etc. and is extracted a large amount of manual feature to portray Deceptive news, achieve
More good result.Different from the feature that traditional-handwork portrays Deceptive news, the method based on deep learning no longer needs to carry out
Complicated Feature Engineering.The social attribute of Deceptive news is regarded as elongated time series by Ma et al., and uses circulation nerve net
Network (RNN) is handled, and recognition accuracy has larger promotion compared with conventional method.Guo et al. thinks Deceptive news event, correlation
There is hierarchical relationships between news and its comment, and propose level attention model to handle this hierarchical relationship, test
Show that this method has ability more outstanding on identification Deceptive news.
Inventor has found that existing method often excessively depends on news itself when carrying out Deceptive news detection research, and
Lack the analysis to Deceptive news this kind news general character, causes recognition accuracy of the existing method in newly generated news difficult
To reach expected horizontal.
Summary of the invention
In view of the above problems, the invention proposes a kind of deep learning Deceptive news detection algorithms guided using style.
This method guides the learning process of deep learning model by the explicit style and features shared using Deceptive news,
Guarantee the available common feature to Deceptive news entirety of model, improves model and newly generating the detection effect in news.
Wherein antisense " explicitly " corresponds to " implicitly ", " implicit " refer to that in model include these information, but these letters
Breath can not be observed in the external world;" explicit " be meant that by it is special use, the protrusion information guides model.
In view of the deficiencies of the prior art, the present invention proposes a kind of bilinearity neural network Deceptive news based on style guidance
Detection method, including:
The newsletter archive that step 1, acquisition are detected to network false news, quantifies the language of the newsletter archive by neural network
It says style and features, obtains the style vector of the newsletter archive, which is inputted into Text character extraction device, obtains the news
The text vector of text;
The style vector sum text vector is inputted bilinearity neural network by step 2, which includes
Bilinear function, for modeling the correlation between the style vector sum text vector, to obtain the wind of the newsletter archive
Lattice-text feature matrix form boot vector using largest score vector in the style-text feature matrix, and by the guidance
Vector is input to full articulamentum, determines the Deceptive news label of the newsletter archive.
The described bilinearity neural network Deceptive news detection method based on style guidance, wherein the step 1 include:
Step 11 converts the newsletter archive to the vector matrix x that vocabulary vector is spliced1:n=x1⊕x2⊕…⊕
xn, wherein ⊕ indicates concatenation, xiIndicate vocabulary vector, x corresponding to i-th of word in the newsletter archive1:nIndicate that length is
The vector matrix of n;
Step 12, text feature extractor are shot and long term memory network, which is input to shot and long term memory
Network obtains the hidden state h of each vocabulary vector in the vector matrixt=H (ht-1,xt), t is less than or equal to n, htIt is t-th
The hidden state of vocabulary vector;
Step 13, by attention mechanism come for each hidden state assignment weightui=tanh
(Wwhi+bw), wherein W*Indicate weight matrix, bwIndicate biasing, αiFor the weight of i-th of hidden state after normalization;
Step 14 passes through the weighted sum weight αiWith the hidden state ht, obtain text vector ftFor text vector.
The bilinearity neural network Deceptive news detection method based on style guidance, wherein obtaining the style-text
The method of eigen matrix is as follows:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
The bilinearity neural network Deceptive news detection method that any one described is guided based on style, the wherein text
Feature extractor is shot and long term memory network or two-way shot and long term memory network.
The bilinearity neural network Deceptive news detection method based on style guidance, wherein step 2 includes: to use
Maximum pond function filters out largest score vector in the style-text feature matrix and forms boot vector.
The invention also provides a kind of bilinearity neural network Deceptive news detection systems based on style guidance, wherein wrapping
It includes:
The newsletter archive that module 1, acquisition are detected to network false news, quantifies the language of the newsletter archive by neural network
It says style and features, obtains the style vector of the newsletter archive, which is inputted into Text character extraction device, obtains the news
The text vector of text;
The style vector sum text vector is inputted bilinearity neural network by module 2, which includes
Bilinear function, for modeling the correlation between the style vector sum text vector, to obtain the wind of the newsletter archive
Lattice-text feature matrix form boot vector using largest score vector in the style-text feature matrix, and by the guidance
Vector is input to full articulamentum, determines the Deceptive news label of the newsletter archive.
The described bilinearity neural network Deceptive news detection system based on style guidance, wherein the module 1 include:
Module 11 converts the newsletter archive to the vector matrix x that vocabulary vector is spliced1:n=x1⊕x2⊕…⊕
xn, wherein ⊕ indicates concatenation, xiIndicate vocabulary vector, x corresponding to i-th of word in the newsletter archive1:nIndicate that length is
The vector matrix of n;
Module 12, text feature extractor are shot and long term memory network, which is input to shot and long term memory
Network obtains the hidden state h of each vocabulary vector in the vector matrixt=H (ht-1,xt), t is less than or equal to n, htIt is t-th
The hidden state of vocabulary vector;
Module 13, by attention mechanism come for each hidden state assignment weightui=tanh
(Wwhi+bw), wherein W*Indicate weight matrix, bwIndicate biasing, αiFor the weight of i-th of hidden state after normalization;
Module 14 passes through the weighted sum weight αiWith the hidden state ht, obtain text vector
ftFor text vector.
The bilinearity neural network Deceptive news detection system based on style guidance, wherein obtaining the style-text
The system of eigen matrix is as follows:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
The bilinearity neural network Deceptive news detection system that any one described is guided based on style, the wherein text
Feature extractor is shot and long term memory network or two-way shot and long term memory network.
The bilinearity neural network Deceptive news detection system based on style guidance, wherein module 2 includes: to use
Maximum pond function filters out largest score vector in the style-text feature matrix and forms boot vector.
As it can be seen from the above scheme the present invention has the advantages that
Due to the proposition using diction feature guidance network learning method, compared with prior art, the present invention
There is higher recognition accuracy in newborn media event.It is previous be not added intervention, it is simple using complicated deep learning
What the method for the feature of model autonomous learning Deceptive news often learnt is the relevant feature of event, these features are difficult to migrate
It uses in newborn event, causes Generalization Capability of the model in newborn event poor.Different from this, the explicit benefit of the present invention
The learning process that deep learning model is guided with the diction feature according to general character in acquirement of expert knowledge Deceptive news, makes
Deep learning model focuses on the common feature of Deceptive news, so that feature acquired in model also has in newborn event
Good migration greatly improves recognition accuracy and the Generalization Capability of model.
Detailed description of the invention
Fig. 1 is the learning framework figure of knowledge elicitation;
Fig. 2 is the bilinearity neural network algorithm flow chart guided based on style.
Specific embodiment
The object of the present invention is to provide a kind of Deceptive news detection method of knowledge elicitation, mainly solve the problems, such as how be
The feature for having more generalization is obtained using the diction feature pilot model of Deceptive news general character, is generated with improving model newly
Detection effect in news.
Key point of the present invention includes:
1, diction quantifies: the expression-form of diction finger speech speech is mainly manifested in vocabulary, grammer, rhetorical device
In equal distribution difference, diction focus on event how to be expressed rather than event content itself.But diction is a pumping
As concept, it need to be quantified according to specific requirements;
2, text feature extract: text feature be model determine news whether be Deceptive news important evidence.In this hair
In bright, extracted using the two-way length memory network (LSTM with Attention Mechanism) with attention mechanism
Text feature, using the vector of extracted feature as the feature representation of newsletter archive;
3, the learning framework of knowledge elicitation: using diction information come the study of correct guidance deep learning model
Journey.It is introduced into what is rationalized using the extracted diction feature of expertise in the learning process of deep learning model
It goes, keeps the learning process of model controllable, so that model focuses on the common feature of Deceptive news and nonspecific single media event
Feature, ensure that generalization of the model in newborn media event.
To allow features described above and effect of the invention that can illustrate more clearly understandable, special embodiment below, and cooperate
Bright book attached drawing is described in detail below.
One, diction quantifies.
Diction is the abstract concept of description language expression-form, usual diction and vocabulary, grammer and rhetoric hand
Method is closely related.But specifically this abstract concept is quantified with mathematical form description language style needs, could be made
It uses in model.Invention defines eight major class, totally 54 features carry out description language style, are defined as follows shown in table 1,
Wherein XBIndicate binary (0/1) feature, XNIndicate numeric type feature.
Table 1:
Diction after quantization is expressed as a string of vectors, such as using table 1, one section of content of text is quantized into a string
This vector, then connect with full articulamentum, obtains style vector so that language by the digital vectors as [0,1,1,0.9,0]
Style information expression is richer, and guides study to Deceptive news detection depth model whereby.
Two, text feature extracts:
Text be model determine news whether be Deceptive news important evidence, model can not directly handle natural language,
Therefore, it is necessary to could handle after being translated into vector matrix.Word2Vec method is used in the present invention in training set corpus
The feature representation of each vocabulary in middle training corpus, and be the matrix that vocabulary vector is spliced by news sentence expression, it indicates
Form is as follows:
x1:n=x1⊕x2⊕…⊕xn
Wherein ⊕ indicates the concatenation of feature, xiIndicate feature vector (word corresponding to i-th of word in the newsletter archive
Remittance vector), x1:nIndicate spliced sentence characteristics vector, length n.
Feature extractor of the two-way shot and long term memory network (LSTM) as text is used in the present invention, LSTM is extensive
Be used in the handling elongated sequence of the task, and machine translation, speech recognition, question answering system etc. application in achieve order
The performance that people attractes attention.LSTM uses word x at each momentiWith the hidden state h of last momentt-1As input, and export current hidden
State ht, Formal Representation are as follows:
ht=H (ht-1,xt)
WhereinIndicate activation primitive.Two-way shot and long term memory network (BiLSTM) solves LSTM can only be in the past
The deficiency that information is obtained in text, by (positive reversed simultaneously using a LSTM to inputting while carrying out forward and reverse input
Use a LSTM), the information of the available current vocabulary context of BiLSTM, so that sentence expression is more abundant.Many institute's weeks
Know, in a sentence and not all word is all of equal importance, wherein usually containing some words than other vocabulary with richer
Information.Therefore, it is less appropriate for being put on an equal footing to all words using identical weight, introduces attention machine in the present invention
(Attention Mechanism) is made to practise different weights for different lexicography, weight shows more greatly the word in sentence more
Important, the sentence after introducing attention mechanism is expressed as follows:
ui=tanh (Wwhi+bw
Wherein W*It indicates weight matrix, is obtained by neural network learning, it is random initial for most starting weight matrix
Change, by constantly learning, reasonable weight matrix can be obtained.bwIndicate biasing, αiTo use softmax function normalization
Weight later, ftIt is expressed as the weighted sum of the hidden vector of all moment BiLSTM outputs.So far text vector has just been got
Final vector expression, and will use this to carry out Deceptive news judgement.
Three, the learning framework of knowledge elicitation:
As shown in Figure 1, knowledge elicitation frame is input with style vector and text vector, wind is obtained using bilinear function
Lattice-text feature matrix, mathematical expression are as follows:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
Bilinear function is proposed to modeling double factor variable, such as " style " and " content ", and has got excellent
Effect.Bilinear function can model the correlation between two variables well, and can capture the phase interaction between variable
With relationship.Using bilinear function processing diction vector and Text eigenvector can be very good capture style and text it
Between mutual response relationship, it can model the response in each dimension of text vector to style and features, i.e., each dimension packet
Style information containing how many degree.This operation be equivalent to using expertise for deep learning model extraction go out feature to
Amount is given a mark, and high score is given in the part with obvious style and features, and no obvious style and features part will give low point
Number.Then, being filtered out using maximum pond function (Max-Pooling) has largest score in style-text feature matrix
Vector forms boot vector (Guided Feature), which indicates to filter out in text the most portion of style and features information
Point.Automatically obtain that text feature is different, which is to be added to expertise from simple utilization deep learning frame
, it is that learning process intervene and instructed using knowledge is artificial, that is to say, that in feature selecting part, both wrapped
Knowledge containing machine: the text feature that deep learning model obtains automatically, and include expertise: expert teaches model which feature
It is really important.This boot vector is connect with full articulamentum to carry out Deceptive news detection, in the present invention, 0 by the present invention
Non- Deceptive news are represented, 1 represents Deceptive news.
The following are system embodiment corresponding with above method embodiment, present embodiment can be mutual with above embodiment
Cooperation is implemented.The relevant technical details mentioned in above embodiment are still effective in the present embodiment, in order to reduce repetition,
Which is not described herein again.Correspondingly, the relevant technical details mentioned in present embodiment are also applicable in above embodiment.
The invention also provides a kind of bilinearity neural network Deceptive news detection systems based on style guidance, wherein wrapping
It includes:
The newsletter archive that module 1, acquisition are detected to network false news, quantifies the language of the newsletter archive by neural network
It says style and features, obtains the style vector of the newsletter archive, which is inputted into Text character extraction device, obtains the news
The text vector of text;
The style vector sum text vector is inputted bilinearity neural network by module 2, which includes
Bilinear function, for modeling the correlation between the style vector sum text vector, to obtain the wind of the newsletter archive
Lattice-text feature matrix form boot vector using largest score vector in the style-text feature matrix, and by the guidance
Vector is input to full articulamentum, determines the Deceptive news label of the newsletter archive.
The described bilinearity neural network Deceptive news detection system based on style guidance, wherein the module 1 include:
Module 11 converts the newsletter archive to the vector matrix x that vocabulary vector is spliced1:n=x1⊕x2⊕…⊕
xn, wherein ⊕ indicates concatenation, xiIndicate vocabulary vector, x corresponding to i-th of word in the newsletter archive1:nIndicate that length is
The vector matrix of n;
Module 12, text feature extractor are shot and long term memory network, which is input to shot and long term memory
Network obtains the hidden state h of each vocabulary vector in the vector matrixt=H (ht-1,xt), t is less than or equal to n, htIt is t-th
The hidden state of vocabulary vector;
Module 13, by attention mechanism come for each hidden state assignment weightui=
tanh(Wwhi+bw), wherein W*Indicate weight matrix, bwIndicate biasing, αiFor the weight of i-th of hidden state after normalization;
Module 14 passes through the weighted sum weight αiWith the hidden state ht, obtain text vector
ftFor text vector.
The bilinearity neural network Deceptive news detection system based on style guidance, wherein obtaining the style-text
The system of eigen matrix is as follows:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
The bilinearity neural network Deceptive news detection system that any one described is guided based on style, the wherein text
Feature extractor is shot and long term memory network or two-way shot and long term memory network.
The bilinearity neural network Deceptive news detection system based on style guidance, wherein module 2 includes: to use
Maximum pond function filters out largest score vector in the style-text feature matrix and forms boot vector.
Claims (10)
1. a kind of bilinearity neural network Deceptive news detection method based on style guidance characterized by comprising
The newsletter archive that step 1, acquisition are detected to network false news, quantifies the language wind of the newsletter archive by neural network
Lattice feature obtains the style vector of the newsletter archive, which is inputted Text character extraction device, obtains the newsletter archive
Text vector;
The style vector sum text vector is inputted bilinearity neural network by step 2, which includes two-wire
Property function, for modeling the correlation between the style vector sum text vector, to obtain style-text of the newsletter archive
Eigen matrix forms boot vector using largest score vector in the style-text feature matrix, and the boot vector is defeated
Enter to determine to full articulamentum the Deceptive news label of the newsletter archive.
2. the bilinearity neural network Deceptive news detection method as described in claim 1 based on style guidance, feature exist
In the step 1 includes:
Step 11 converts the newsletter archive to the vector matrix x that vocabulary vector is spliced1:n=x1⊕x2⊕…⊕xn,
Middle ⊕ indicates concatenation, xiIndicate vocabulary vector, x corresponding to i-th of word in the newsletter archive1:nIndicate that length is being somebody's turn to do for n
Vector matrix;
Step 12, text feature extractor are shot and long term memory network, which is input to the shot and long term memory network,
Obtain the hidden state h of each vocabulary vector in the vector matrixt=H (ht-1,xt), t is less than or equal to n, htFor t-th of vocabulary to
The hidden state of amount;
Step 13, by attention mechanism come for each hidden state assignment weightui=tanh (Wwhi
+bw), wherein W*Indicate weight matrix, bwIndicate biasing, αiFor the weight of i-th of hidden state after normalization;
Step 14 passes through the weighted sum weight αiWith the hidden state ht, obtain text vector ftFor
Text vector.
3. the bilinearity neural network Deceptive news detection method as described in claim 1 based on style guidance, feature exist
In it is as follows to obtain the style-text feature matrix method:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
4. any one bilinearity neural network Deceptive news detection method based on style guidance as described in claim 1,
It is characterized in that, text feature extractor is shot and long term memory network or two-way shot and long term memory network.
5. the bilinearity neural network Deceptive news detection method as described in claim 1 based on style guidance, feature exist
In step 2 includes: to filter out largest score vector composition guidance in the style-text feature matrix using maximum pond function
Vector.
6. a kind of bilinearity neural network Deceptive news detection system based on style guidance characterized by comprising
The newsletter archive that module 1, acquisition are detected to network false news, quantifies the language wind of the newsletter archive by neural network
Lattice feature obtains the style vector of the newsletter archive, which is inputted Text character extraction device, obtains the newsletter archive
Text vector;
The style vector sum text vector is inputted bilinearity neural network by module 2, which includes two-wire
Property function, for modeling the correlation between the style vector sum text vector, to obtain style-text of the newsletter archive
Eigen matrix forms boot vector using largest score vector in the style-text feature matrix, and the boot vector is defeated
Enter to determine to full articulamentum the Deceptive news label of the newsletter archive.
7. the bilinearity neural network Deceptive news detection system as claimed in claim 6 based on style guidance, feature exist
In the module 1 includes:
Module 11 converts the newsletter archive to the vector matrix x that vocabulary vector is spliced1:n=x1⊕x2⊕…⊕xn,
Middle ⊕ indicates concatenation, xiIndicate vocabulary vector, x corresponding to i-th of word in the newsletter archive1:nIndicate that length is being somebody's turn to do for n
Vector matrix;
Module 12, text feature extractor are shot and long term memory network, which is input to the shot and long term memory network,
Obtain the hidden state h of each vocabulary vector in the vector matrixt=H (ht-1,xt), t is less than or equal to n, htFor t-th of vocabulary to
The hidden state of amount;
Module 13, by attention mechanism come for each hidden state assignment weightui=tanh (Wwhi
+bw), wherein W*Indicate weight matrix, bwIndicate biasing, αiFor the weight of i-th of hidden state after normalization;
Module 14 passes through the weighted sum weight αiWith the hidden state ht, obtain text vector ftFor
Text vector.
8. the bilinearity neural network Deceptive news detection system as claimed in claim 6 based on style guidance, feature exist
In it is as follows to obtain the style-text feature matrix system:
fsFor the style vector, ftFor text vector, it is f that B, which is the bilinear function,bThe style-text feature matrix.
9. any one bilinearity neural network Deceptive news detection system based on style guidance as claimed in claim 6,
It is characterized in that, text feature extractor is shot and long term memory network or two-way shot and long term memory network.
10. the bilinearity neural network Deceptive news detection system as claimed in claim 6 based on style guidance, feature exist
In module 2 includes: to filter out largest score vector composition guidance in the style-text feature matrix using maximum pond function
Vector.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910341056.6A CN110210016B (en) | 2019-04-25 | 2019-04-25 | Method and system for detecting false news of bilinear neural network based on style guidance |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910341056.6A CN110210016B (en) | 2019-04-25 | 2019-04-25 | Method and system for detecting false news of bilinear neural network based on style guidance |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110210016A true CN110210016A (en) | 2019-09-06 |
CN110210016B CN110210016B (en) | 2021-06-04 |
Family
ID=67786471
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910341056.6A Active CN110210016B (en) | 2019-04-25 | 2019-04-25 | Method and system for detecting false news of bilinear neural network based on style guidance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110210016B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110889430A (en) * | 2019-10-24 | 2020-03-17 | 中国科学院计算技术研究所 | News image detection method, system and device based on multi-domain visual features |
CN110968672A (en) * | 2019-12-03 | 2020-04-07 | 北京工商大学 | False public opinion identification method for food safety based on neural network |
CN111160040A (en) * | 2019-12-26 | 2020-05-15 | 西安交通大学 | Information reliability evaluation system and method based on multi-scale gating equilibrium interaction fusion network |
CN111221941A (en) * | 2020-01-10 | 2020-06-02 | 复旦大学 | Social media rumor identification algorithm based on text content and literary style |
CN111553146A (en) * | 2020-05-09 | 2020-08-18 | 杭州中科睿鉴科技有限公司 | News writing style modeling method, writing style-influence analysis method and news quality evaluation method |
CN111639252A (en) * | 2020-05-18 | 2020-09-08 | 华中科技大学 | False news identification method based on news-comment relevance analysis |
CN112199606A (en) * | 2020-10-30 | 2021-01-08 | 福州大学 | Social media-oriented rumor detection system based on hierarchical user representation |
CN112711951A (en) * | 2021-01-05 | 2021-04-27 | 西安交通大学 | Induction consciousness-based false news interpretability detection system and method |
CN113177110A (en) * | 2021-05-28 | 2021-07-27 | 中国人民解放军国防科技大学 | False news detection method and device, computer equipment and storage medium |
CN113312479A (en) * | 2021-05-13 | 2021-08-27 | 杭州中科睿鉴科技有限公司 | Cross-domain false news detection method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106354710A (en) * | 2016-08-18 | 2017-01-25 | 清华大学 | Neural network relation extracting method |
CN107066464A (en) * | 2016-01-13 | 2017-08-18 | 奥多比公司 | Semantic Natural Language Vector Space |
CN108776807A (en) * | 2018-05-18 | 2018-11-09 | 复旦大学 | It is a kind of based on can the double branch neural networks of skip floor image thickness grain-size classification method |
CN108875827A (en) * | 2018-06-15 | 2018-11-23 | 广州深域信息科技有限公司 | A kind of method and system of fine granularity image classification |
WO2019025895A1 (en) * | 2017-07-31 | 2019-02-07 | 株式会社半導体エネルギー研究所 | Display device |
CN109344391A (en) * | 2018-08-23 | 2019-02-15 | 昆明理工大学 | Multiple features fusion Chinese newsletter archive abstraction generating method neural network based |
-
2019
- 2019-04-25 CN CN201910341056.6A patent/CN110210016B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107066464A (en) * | 2016-01-13 | 2017-08-18 | 奥多比公司 | Semantic Natural Language Vector Space |
CN106354710A (en) * | 2016-08-18 | 2017-01-25 | 清华大学 | Neural network relation extracting method |
WO2019025895A1 (en) * | 2017-07-31 | 2019-02-07 | 株式会社半導体エネルギー研究所 | Display device |
CN108776807A (en) * | 2018-05-18 | 2018-11-09 | 复旦大学 | It is a kind of based on can the double branch neural networks of skip floor image thickness grain-size classification method |
CN108875827A (en) * | 2018-06-15 | 2018-11-23 | 广州深域信息科技有限公司 | A kind of method and system of fine granularity image classification |
CN109344391A (en) * | 2018-08-23 | 2019-02-15 | 昆明理工大学 | Multiple features fusion Chinese newsletter archive abstraction generating method neural network based |
Non-Patent Citations (4)
Title |
---|
刘飞龙 等: "基于双线性函数注意力Bi-LSTM模型的机器阅读理解", 《计算机科学》 * |
段大高 等: "基于神经网络的微博虚假消息识别模型", 《信息网络安全》 * |
王恰: "基于 Attention Bi-LSTM 的文本分类方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
金志威 等: "融合多模态特征的社会多媒体谣言检测技术研究", 《南京信息工程大学学报》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110889430A (en) * | 2019-10-24 | 2020-03-17 | 中国科学院计算技术研究所 | News image detection method, system and device based on multi-domain visual features |
CN110968672A (en) * | 2019-12-03 | 2020-04-07 | 北京工商大学 | False public opinion identification method for food safety based on neural network |
CN110968672B (en) * | 2019-12-03 | 2022-06-10 | 北京工商大学 | False public opinion identification method for food safety based on neural network |
CN111160040A (en) * | 2019-12-26 | 2020-05-15 | 西安交通大学 | Information reliability evaluation system and method based on multi-scale gating equilibrium interaction fusion network |
CN111221941A (en) * | 2020-01-10 | 2020-06-02 | 复旦大学 | Social media rumor identification algorithm based on text content and literary style |
CN111221941B (en) * | 2020-01-10 | 2023-12-26 | 复旦大学 | Social media rumor authentication algorithm based on text content and line style |
CN111553146A (en) * | 2020-05-09 | 2020-08-18 | 杭州中科睿鉴科技有限公司 | News writing style modeling method, writing style-influence analysis method and news quality evaluation method |
CN111639252A (en) * | 2020-05-18 | 2020-09-08 | 华中科技大学 | False news identification method based on news-comment relevance analysis |
CN112199606A (en) * | 2020-10-30 | 2021-01-08 | 福州大学 | Social media-oriented rumor detection system based on hierarchical user representation |
CN112711951A (en) * | 2021-01-05 | 2021-04-27 | 西安交通大学 | Induction consciousness-based false news interpretability detection system and method |
CN113312479A (en) * | 2021-05-13 | 2021-08-27 | 杭州中科睿鉴科技有限公司 | Cross-domain false news detection method |
CN113177110A (en) * | 2021-05-28 | 2021-07-27 | 中国人民解放军国防科技大学 | False news detection method and device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110210016B (en) | 2021-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110210016A (en) | Bilinearity neural network Deceptive news detection method and system based on style guidance | |
CN110134771A (en) | A kind of implementation method based on more attention mechanism converged network question answering systems | |
CN107766324B (en) | Text consistency analysis method based on deep neural network | |
CN108549658A (en) | A kind of deep learning video answering method and system based on the upper attention mechanism of syntactic analysis tree | |
CN105631468B (en) | A kind of picture based on RNN describes automatic generation method | |
CN107818164A (en) | A kind of intelligent answer method and its system | |
CN107169035A (en) | A kind of file classification method for mixing shot and long term memory network and convolutional neural networks | |
CN108804495A (en) | A kind of Method for Automatic Text Summarization semantic based on enhancing | |
CN111310439B (en) | Intelligent semantic matching method and device based on depth feature dimension changing mechanism | |
CN111325028A (en) | Intelligent semantic matching method and device based on deep hierarchical coding | |
CN111143672B (en) | Knowledge graph-based professional speciality scholars recommendation method | |
CN109992788B (en) | Deep text matching method and device based on unregistered word processing | |
CN107679225A (en) | A kind of reply generation method based on keyword | |
CN112364161A (en) | Microblog theme mining method based on dynamic behaviors of heterogeneous social media users | |
CN113011196A (en) | Concept-enhanced representation and one-way attention-containing subjective question automatic scoring neural network model | |
CN109740164A (en) | Based on the matched electric power defect rank recognition methods of deep semantic | |
CN111444328B (en) | Natural language automatic prediction and inference method with interpretation generation | |
CN114297399A (en) | Knowledge graph generation method, knowledge graph generation system, storage medium and electronic equipment | |
CN114282592A (en) | Deep learning-based industry text matching model method and device | |
CN116522165B (en) | Public opinion text matching system and method based on twin structure | |
CN110826341A (en) | Semantic similarity calculation method based on seq2seq model | |
CN116521886A (en) | Deep learning-based education field discipline knowledge graph construction method and device | |
CN113378574B (en) | KGANN-based named entity identification method | |
CN115346158A (en) | Video description method based on coherence attention mechanism and double-stream decoder | |
CN104965818A (en) | Project name entity identification method and system based on self-learning rules |
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 |