AU2020103810A4 - A method for detecting fake news using grammatic transformation on neural network computer readable medium - Google Patents
A method for detecting fake news using grammatic transformation on neural network computer readable medium Download PDFInfo
- Publication number
- AU2020103810A4 AU2020103810A4 AU2020103810A AU2020103810A AU2020103810A4 AU 2020103810 A4 AU2020103810 A4 AU 2020103810A4 AU 2020103810 A AU2020103810 A AU 2020103810A AU 2020103810 A AU2020103810 A AU 2020103810A AU 2020103810 A4 AU2020103810 A4 AU 2020103810A4
- Authority
- AU
- Australia
- Prior art keywords
- sentence
- news
- vector
- neural network
- proposition
- 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.)
- Ceased
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 title claims abstract description 39
- 230000009466 transformation Effects 0.000 title claims abstract description 35
- 239000013598 vector Substances 0.000 claims abstract description 122
- 238000003058 natural language processing Methods 0.000 claims abstract description 18
- 238000010586 diagram Methods 0.000 claims abstract description 5
- 230000004048 modification Effects 0.000 claims abstract description 4
- 238000012986 modification Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 16
- 230000007246 mechanism Effects 0.000 claims description 9
- 230000006403 short-term memory Effects 0.000 claims description 8
- 230000001537 neural effect Effects 0.000 claims 1
- 238000013135 deep learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 241000282412 Homo Species 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/253—Grammatical analysis; Style critique
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
A method for detecting fake news using grammatic transformation on
neural network computer readable medium
Present invention discloses a method for searching fake news using grammar
5 transformation on a neural network, and a recording medium and apparatus for
performing the same. The fake news search method using grammar transformation on a
neural network is the step of embedding a proposition sentence, which is determined to
be true or false by a news sentence, as a word vector, and inputting the word vector into a
predetermined natural language processing neural network (context) generating a vector,
10 generating a candidate sentence that is a sentence having the same meaning as the
proposition sentence but different grammar by inputting the context vector into a
predetermined natural language processing neural network, and comparing the candidate
sentence with the news sentence And determining whether the proposition sentence is
true or false and searching whether the news sentence corresponds to fake news.
15
21
Application Number: Page 1 of 2
Word Context
Embedding Generating Matching Reasoning
Unit Unit Unit Unit
10 30 50 70
FIG. 1 is a block diagram of an apparatus for searching fake news using grammar
modification on a neural network
Description
Application Number: Page 1 of 2
Context Word Embedding Generating Matching Reasoning Unit Unit Unit Unit
10 30 50 70
FIG. 1 is a block diagram of an apparatus for searching fake news using grammar modification on a neural network
Australian Government
IP Australia
Innovation Patent Application Australia Patent Office
1. TITLE OF THE INVENTION A METHOD FOR DETECTING FAKE NEWS USING GRAMMATIC TRANSFORMATION ON NEURAL NETWORK COMPUTER READABLE MEDIUM
2. APPLICANTS (S) NAME NATIONALITY ADDRESS RAJENDER KUMAR INDIAN DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING, PANIPAT INSTITUTE OF ENGINEERING AND TECHNOLOGY, SAMALKHA, PANIPAT.
GOEL COMMUNICATION, FACULTY OF LIBERAL ARTS AND MEDIA STUDIES. J.C. BOSE UNIVERSITY OF SCIENCE AND TECHNOLOGY , YMCA, FARIDABAD DR. BASANT AGARWAL DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING, INDIAN INSTITUTE OF INFORMATION TECHNOLOGY, KOTA (MNIT CAMPUS JAIPUR-302017) DR. PRIYANKA HARJULE DEPARTMENT OF MATHEMATICS, INDIAN INSTITUTE OF INFORMATION TECHNOLOGY, KOTA (MNIT CAMPUS JAIPUR-302017)
3. PREAMBLE TO THE DESCRIPTION
The following specification particularly describes the invention and the manner in which it is to be performed
Field of the Invention
[OOO]The present invention relates to a method for searching fake news using grammar transformation on a neural network, recording medium and device for performing the same.More particularly, to generate a sentence based on deep learning, and to determine whether the news is authentic or not.
Background of the Invention
[0002]With the advent of the mobile era in which high-speed mobile communication networks and smart phones are widely spread, the use of various social network services (SNS) is rapidly increasing. In particular, in recent years, the use of SNS such as blog, KakaoTalk, Line, Facebook, Twitter, Instagram, and Tumblr has rapidly increased. Meanwhile, the delivery of information or news through various SNS is also increasing explosively.
[0003]However, whenever there is a political event such as an election, misinformation or news is spread through various SNSs. In addition, even more seriously, there are frequent cases of intentionally distributing false information or news through various SNS for a specific purpose. For this reason, the delivery of information or news through SNS may become an important social issue in the future.
[0004]Meanwhile, fake news, which is news intentionally manipulated with a specific purpose as above, is mostly searched by humans, and its authenticity is determined.Since this requires a lot of time and effort, it does not require human judgment, and a new fake news search model that can select fake news in real time is needed.
Summary of the Invention
[0005]The present invention is a fake news search method using a grammar transformation on a neural network that generates a candidate sentence in which the grammar of a proposition sentence is modified based on deep learning, and compares the candidate sentence with the news sentence to determine the authenticity of the news, a recording medium and a device for performing this are provided.
[0006]The method for searching fake news using grammar transformation on a neural network according to the present invention includes the steps of embedding a proposition sentence, which is determined to be true or false by a news sentence, into a word vector, and the word vector is a predetermined natural language . Generating a context vector by inputting the context vector into a predetermined natural language processing neural network to generate a candidate sentence that has the same meaning as the proposition sentence but a different grammar, and the candidate sentence and the comparing news sentences to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news.
[0007]On the other hand, the step of embedding a proposition sentence, which is determined to be true or false by the news sentence, into a word vector, comprises constructing the proposition sentence using a one-hot encoding method. It may include converting each word into a vector.
[0008]In addition, the step of generating the context vector by inputting the word vector into a predetermined natural language processing neural network includes the step of generating the context vector by inputting the word vector into a long short term memory (LSTM) neural network.
[0009]In addition, the step of generating a candidate sentence, which is a sentence having the same meaning as the proposition sentence but different grammar, by inputting the context vector into a predetermined natural language processing neural network, the attention mechanism of a sequence-to-sequence learning model Generating an attention vector by calculating a weighted sum of the context vector according to the method, comparing the attention vector and a hidden state vector generated when the context vector is generated to obtain a matching vector. It may include predicting and generating the candidate sentence by inputting the matching vector into a Long Short Term Memory (LSTM) neural network.
[0010]In addition, the step of comparing the candidate sentence with the news sentence to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news may include a sequence of a vector corresponding to each word constituting the candidate sentence. Selecting and outputting a given number of words by inputting them to the softmax function of the sequence to sequence learning model, and combining the output words of the softmax function using a beam search decoder Generating a final candidate sentence and comparing the final candidate sentence with the news sentence to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news.
[0011]ln addition, the step of comparing the final candidate sentence with the news sentence to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news includes embedding the final candidate sentence and the news sentence, and calculating a cosine similarity of the embedded final candidate sentence and the news sentence, and determining true or false of the proposition sentence by the news sentence by using the cosine similarity.
[0012]In addition, it may be a computer-readable recording medium in which a computer program is recorded for performing a method of searching for fake news using grammar transformation on the neural network.
[0013]On the other hand, the apparatus for searching fake news using grammar transformation on a neural network according to the present invention includes a word embedding unit for embedding a proposition sentence, which is determined to be true or false by a news sentence, as a word vector. A context generator that generates a context vector by inputting it into a natural language processing neural network of, and a matching that generates a candidate sentence that is a sentence having the same meaning as the proposition sentence but different grammar by inputting the context vector into a predetermined natural language processing neural network And a reasoning unit that compares the negative and the candidate sentence with the news sentence to determine whether the proposition sentence is true or false and searches whether the news sentence corresponds to fake news.
[0014]Meanwhile, the word embedding unit may convert each word constituting the proposition sentence into a vector using a one-hot encoding method.
[0015]In addition, the context generator may generate the context vector by inputting the word vector into a long short term memory (LSTM) neural network.
[0016]In addition, the matching unit generates an attention vector by calculating a weighted sum of the context vector according to an attention mechanism of a sequence to-sequence learning model, and generates an attention vector, wherein the matching vector is predicted by comparing a hidden state vector generated when a context vector is generated, and the candidate sentence may be generated by inputting the matching vector into a long short term memory (LSTM) neural network.
[0017]In addition, the inference unit selects and outputs a given number of words by inputting a vector corresponding to each word constituting the candidate sentence into a softmax function of a sequence-to-sequence learning model. , Using a beam search decoder, a final candidate sentence is generated by combining the output words of the softmax function, and the final candidate sentence is compared with the news sentence to determine true or false of the proposition sentence, and the news You can search if the sentence corresponds to fake news.
[0018]In addition, the inference unit embeds the final candidate sentence and the news sentence, calculates a cosine similarity of the embedded final candidate sentence and the news sentence, and uses the cosine similarity to determine the proposition sentence by the news sentence. It can determine true or false.
[0019]According to the present invention, human judgment is not required, and fake news can be screened and processed in real time.
In addition, by selecting fake news, it is possible to prevent the spread of fake news, thereby preventing social confusion that may be caused by fake news.
Brief Description of the Drawings
[0020]FIG.1 is a block diagram of an apparatus for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention.
FIG. 2 is a flowchart of a method for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention.
Detailed Description
[0021] Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, and only these embodiments make the disclosure of the present invention complete, and the general knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to those who have it, and the invention is only defined by the scope of the claims. The same reference numerals refer to the same elements throughout the specification.
[0022]The terms used in the present specification are for describing exemplary embodiments and are not intended to limit the present invention. In this specification, the singular form also includes the plural form unless specifically stated in the phrase. As used herein, "comprises" and/or "comprising" do not exclude the presence or addition of one or more other elements, steps, and actions to the mentioned elements, steps, and actions.
[0024] FIG. 1 is a block diagram of an apparatus for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention.
[0025] Referring to FIG. 1, a fake news search apparatus 1 using a grammar transformation on a neural network according to an embodiment of the present invention includes a word embedding unit 10, a context generating unit 30, a matching unit 50, and a reasoning unit. (70) may be included.
[0026]The fake news search device 1 using the grammar transformation on a neural network to determine the true or false of a news sentence, a candidate having modified the grammar of a proposition sentence based on deep learning. You can create a sentence and use it to determine whether the news is authentic or not. Therefore, the fake news search apparatus 1 using grammar transformation on a neural network according to an embodiment of the present invention can present a fast and efficient fake news search model without human intervention.
[0027]The fake news search device 1 using grammar transformation on a neural network according to an embodiment of the present invention may be executed by installing software (application) for searching fake news, and a word embedding unit 10 and a context generating unit 30 ), the matching unit 50 and the reasoning unit 70 may be controlled by software for searching for fake news.
[0028]The configuration of the word embedding unit 10, the context generation unit 30, the matching unit 50, and the inference unit 70 may be formed of an integrated module or may be formed of one or more modules. However, on the contrary, each component may be formed as a separate module.
[0029]The apparatus 1 for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention may have mobility or may be fixed. The fake news search apparatus 1 using grammar transformation on a neural network according to an embodiment of the present invention may be in the form of a computer, a server, or an engine, and may be a device or an apparatus. , Terminal, UE (user equipment), MS (mobile station), MT (mobile terminal), UT (user terminal), SS (subscriber station), wireless device (wireless device), PDA (personal digital assistant), It may be referred to by other terms such as wireless modem and handheld device.
[0030]Hereinafter, each configuration of the fake news search apparatus 1 using a grammar transformation on a neural network according to an embodiment of the present invention shown in FIG. 1 will be described in detail.
[0031] The word embedding unit 10 may embed a proposition sentence, which is determined to be true or false from a news sentence, as a word vector.
[0032] Here, the proposition sentence is a sentence for determining the authenticity of the news subject to be searched in this embodiment. For example, when the proposition sentence is determined to be false by the news sentence, the news may be determined as fake news.
[0033]The word embedding unit 10 may generate a word vector from a proposition sentence through embedding. The word embedding unit 10 may convert each word constituting the proposition sentence into a vector by adopting a one-hot encoding method among the embedding methods. This is because in the present embodiment, candidate sentences that have the same meaning as the proposition sentence but different grammar sentences are generated using neural network grammar transformation, but it is impossible to input the characters themselves into the neural network algorithm.
[0034] Meanwhile, in the present embodiment, a sequence-to-sequence learning model based on a long short term memory (LSTM) neural network may be used to modify the neural network grammar. The context generation unit 30, the matching unit 50, and the inference unit 70, which will be described later, configure each layer of a sequence-to sequence learning model based on an LSTM neural network to perform fake news search.
[0035] The LSTM neural network is a structure in which a cell state is added to the hidden state of a recurrent neural network (RNN) neural network, which is widely used to process the flow between letters and words.
The RNN neural network is a type of artificial neural network in which hidden nodes of the hidden state are connected by directional edges to form a circulatory structure. The cell state added to the LSTM neural network can affect the next cell using a formula that allows the gradient to propagate relatively well even if the state elapses for a long time.
[0036]The sequence-to-sequence learning model is a deep learning algorithm implemented using an LSTM neural network, and it is easy to process data where the dimensions of the input and the output are not fixed, such as text. The sequence-to sequence learning model can receive a source sequence and a target sequence as inputs, and these inputs can be processed by dividing into two parts: an encoder and a decoder. The encoder can convert the source sequence into a vector of a fixed size, and the decoder can convert the vector of the encoder into a target sequence. In this case, in the process of predicting the target sequence, the decoder can predict by putting the predicted result immediately before as an input of the next step. In order to increase the accuracy of sequence prediction in a sequence-to-sequence learning model, an attention mechanism that gives an impact to important words may be used.
[0037]In addition, the sequence-to-sequence learning model includes a softmax layer and finally decodes words to generate sentences. At this time, sentences composed of words with high probability are output using a decoder. can do. Although various types of such decoders have been proposed, a sequence-to-sequence learning model applied to the present embodiment may be implemented including a beam search decoder. The beam search decoder may generate a sentence of an optimal combination by generating a given number of words with the highest probability according to the beam size.
[0038]The context generator 30 may generate a context vector by inputting a word vector into a predetermined natural language processing neural network. Here, the natural language processing neural network may be an LSTM neural network as described above.
[0039]The matching unit 50 may generate candidate sentences that are sentences having the same meaning as the proposition sentence but different grammar using the context vector.
[0040]The matching unit 50 may apply the attention mechanism of the sequence-to sequence learning model as described above so as to increase the accuracy of generating candidate sentences.
[0041]That is, the matching unit 50 may generate the attention vector from the context vector using the attention mechanism of the sequence-to-sequence learning model. The matching unit 50 may generate an attention vector that has an impact on an important word by calculating a weighted sum of the context vectors according to the attention mechanism.
[0042]The matching unit 50 may predict the matching vector by performing a matching operation comparing the attention vector and the hidden state vector generated when the context vector is generated by the context generating unit 30.As described above, the context generator 30 generates a context vector using the LSTM neural network, and a hidden state vector may be generated from the hidden state of the LSTM neural network. The matching unit 50 may predict the matching vector by performing a matching operation comparing the attention vector and the hidden state vector.
[0043]The matching unit 50 may generate a candidate sentence by inputting the matching vector into a predetermined natural language processing neural network. Here, the natural language processing neural network may be an LSTM neural network as described above.
[0044]The reasoning unit 70 may compare the candidate sentence with the news sentence to determine whether the proposition sentence is true or false, and search whether the news sentence corresponds to fake news.
[0045]The reasoning unit 70 compares the candidate sentences of the optimal combination with the news sentences to increase the accuracy of selecting fake news, so that the softmax layer of the sequence-to-sequence learning model described above is Can be used.
[0046]That is, the inference unit 70 may select and output a given number of words by inputting a candidate sentence into a softmax function of a sequence-to-sequence learning model. Here, the candidate sentence may correspond to a concatenated aggregated matching vector that is an output of the matching unit 50, and the inference unit 70 generates a total matching vector corresponding to each word constituting the candidate sentence. By inputting into the softmax function, a given number of words can be selected and output. In this case, the output word may be a word having a high probability of appearing in a news sentence.
[0047]The inference unit 70 may generate a final candidate sentence by inputting the output of the softmax function to the beam search decoder. As described above, the beam search decoder may generate sentences by composing words that are outputs of a softmax function into a combination with a degree of completion.
[0048]The inference unit 70 may generate a candidate sentence with more completeness by configuring a softmax layer of a sequence-to-sequence learning model as described above.
[0049]The inference unit 70 may compare the news sentence and the candidate sentence, which is a sentence in which the grammar of the proposition sentence is modified, in order to determine the true or false proposition sentence by the news sentence.
[0050]To this end, the inference unit 70 may embed a news sentence and a candidate sentence that is an output of the beam search decoder. For example, the inference unit 70 may embed news sentences and candidate sentences using the Doc2Vec model and convert them into vectors. In this case, there may be a plurality of candidate sentences that are outputs of the beam search decoder, and in this case, the inference unit 70 may collect a plurality of candidate sentences to form a sentence group.
[0051]The inference unit 70 may calculate the cosine similarity of the embedded news sentence and the candidate sentence. The reasoning unit 70 may determine true or false of a proposition sentence based on a news sentence using the cosine similarity. For example, the cosine similarity may have a value between 0 and 1, and the inference unit 70 may determine the proposition sentence to be true when the cosine similarity is 0.5 or more.
[0052]When a paragraph group is formed, the inference unit 70 may calculate both the cosine similarity of each candidate sentence of the paragraph group and the news sentence. The inference unit 70 may determine the true or false of the proposition sentence by the news sentence by using the highest cosine similarity among the candidate sentences of the paragraph group and the cosine similarity of the news sentence.
[0053]When the proposition sentence by the news sentence is determined to be true, the reasoning unit 70 may select the news as being true, and when the proposition sentence by the news sentence is determined to be false, the news is also considered to be fake news. Can be selected.
[0054]In this way, the fake news search apparatus 1 using grammar transformation on a neural network according to an embodiment of the present invention constructs an LSTM-based sequence-to-sequence learning model, and has the same meaning as the proposition sentence. However, it is possible to select fake news by creating candidate sentences, which are sentences with different grammar, and comparing the candidate sentences with news sentences.
[0055]The apparatus 1for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention does not require human judgment and can process fake news in real time.
[0057]The present invention illustrating an LSTM-based sequence-to-sequence learning model constructed by a fake news search apparatus using a grammar transformation on a neural network according to an embodiment of the present invention shown in FIG. 1.
[0058]The LSTM-based sequence-to-sequence learning model includes a word embedding layer, a context generation layer, a matching layer, and an inference layer( Inference Layer).
[0059]The word embedding unit 10 configures a word embedding layer, and embeds a proposition sentence, which is determined to be true or false from a news sentence, as a word vector.
[0060]The context generator 30 configures a context generation layer, and may generate a context vector by inputting a word vector into an LSTM neural network.
[0061]The matching unit 50 may configure a matching layer, and may generate a matching vector by performing a matching operation on a context vector.At this time, the matching unit 50 generates an attention vector by inputting the context vector into the attention function, and performs a matching operation of comparing the attention vector with the hidden state vector of the LSTM neural network included in the context generation layer. You can create a matching vector. In addition, the matching unit 50 may input the matching vector into the LSTM neural network to generate an associated sum matching vector corresponding to a candidate sentence that is a sentence having the same meaning as the proposition sentence but different grammar.
[0062]The inference unit 70 configures an inference layer, obtains a sum matching vector from an associated sum matching vector, and inputs it to a softmax function to select and output a given number of words. In addition, the inference unit 70 may generate a plurality of candidate sentences by inputting the output of the softmax function to the beam search decoder. The reasoning unit 70 may compare a plurality of candidate sentences with news sentences to determine the true or false proposition sentence based on the news sentence, and select fake news according to the result.
[0064]Hereinafter, a method of searching for fake news using grammar transformation on a neural network according to an embodiment of the present invention will be described.
[0065]Fig.2 is a flowchart of a method for searching fake news using grammar transformation on a neural network according to an embodiment of the present invention.
[0066]The fake news search method using the grammar transformation on the neural network according to an embodiment of the present invention may be executed in substantially the same configuration as the fake news search apparatus 1 using the grammar transformation on the neural network shown in FIG. 1.Accordingly, the same components as those of the fake news search apparatus 1 using grammar transformation on the neural network shown in FIG. 1 are given the same reference numerals, and repeated descriptions are omitted.
[0067]Referring to FIG. 2, the word embedding unit 10 may embed a proposition sentence as a word vector (S100).
[0068]The proposition sentence corresponds to a sentence to be determined to be true or false by the news sentence. The word embedding unit 10 may convert each word constituting the proposition sentence into a vector by adopting a one-hot encoding method among the embedding methods.
[0069]The context generator 30 may generate a context vector by inputting a word vector into the neural network (S200).
[0070]Here, the neural network is a natural language processing neural network, and may be an LSTM neural network.
[0071]The matching unit 50 may generate a candidate sentence by inputting the context vector into the neural network (S300).
[0072]Candidate sentences correspond to sentences with the same meaning as proposition sentences but different grammar.
[0073]The matching unit 50 may generate an attention vector from a context vector using an attention mechanism of a sequence-to-sequence learning model. The matching unit 50 may generate an attention vector that has an impact on an important word by calculating a weighted sum of the context vectors according to the attention mechanism.
[0074]The matching unit 50 may predict the matching vector by performing a matching operation comparing the hidden state vector generated when the attention vector and the context vector are generated.
[0075]The matching unit 50 may generate a candidate sentence by inputting the matching vector into the LSTM neural network.
[0076]The reasoning unit 70 may search for fake news by comparing the candidate sentence with the news sentence (S400).
[0077]The reasoning unit 70 may compare the candidate sentence with the news sentence to determine whether the proposition sentence is true or false, and search whether the news sentence corresponds to fake news.
[0078]The inference unit 70 may select and output a given number of words by inputting a candidate sentence into a softmax function of a sequence-to-sequence learning model.Here, the candidate sentence may correspond to a concatenated aggregated matching vector that is an output of the matching unit 50, and the inference unit 70 generates a total matching vector corresponding to each word constituting the candidate sentence. By inputting into the softmax function, a given number of words can be selected and output.In this case, the output word may be a word having a high probability of appearing in a news sentence.
[0079]The inference unit 70 may generate a final candidate sentence by inputting the output of the softmax function to the beam search decoder. As described above, the beam search decoder may generate sentences by composing words that are outputs of the softmax sum into a combination with a degree of completion.
[0080]The inference unit 70 may embed a news sentence and a candidate sentence that is an output of the beam search decoder. For example, the reasoning unit 70 may embed a news sentence and a candidate sentence using the Doc2Vec model. In this case, there may be a plurality of candidate sentences that are outputs of the beam search decoder, and in this case, the inference unit 70 may collect a plurality of candidate sentences to form a sentence group.
[0081]The inference unit 70 may calculate the cosine similarity of the embedded news sentence and the candidate sentence. The reasoning unit 70 may determine true or false of a proposition sentence based on a news sentence using the cosine similarity. For example, the cosine similarity may have a value between 0 and 1, and the inference unit 70 may determine the proposition sentence to be true when the cosine similarity is 0.5 or more.
[0082]When a paragraph group is formed, the inference unit 70 may calculate both the cosine similarity of each candidate sentence of the paragraph group and the news sentence.
The inference unit 70 may determine the true or false of the proposition sentence by the news sentence by using the highest cosine similarity among the candidate sentences of the paragraph group and the cosine similarity of the news sentence.
[0084]The fake news search method using the grammar transformation on the neural network of the present invention can be implemented in the form of program instructions that can be executed through various computer components and recorded in a computer readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.
[0085]The program instructions recorded on the computer-readable recording medium may be specially designed and constructed for the present invention, and may be known and usable to those skilled in the computer software field.
[0086]Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magnetic-optical media such as floptical disks. media), and a hardware device specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like.
[0087] Examples of the program instructions include not only machine language codes such as those produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device can be configured to operate as one or more software modules to perform the processing according to the present invention, and vice versa.
[0088]The embodiments of the present invention have been described above with reference to the accompanying drawings, but those of ordinary skill in the art to which the present invention pertains can be implemented in other specific forms without changing the technical spirit or essential features. You can understand. Therefore, it should be understood that the embodiments described above are illustrative and non limiting in all respects.
Claims (9)
1. A method for detecting fake news using grammatic transformation on neural network computer readable medium, wherein the method comprises the steps of:
Embedding a proposition sentence, which is determined to be true or false by a news sentence, into a word vector; Inputting the word vector into a predetermined natural language processing neural network to generate a context vector;
Generating a candidate sentence having the same meaning as the proposition sentence but different grammar by inputting a vector into a predetermined natural language processing neural network; and
comparing the candidate sentence with the news sentence to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news.
2. The method of claim 1, wherein embedding a proposition sentence, which is determined to be true or false by the news sentence, into a word vector, comprises: the proposition using a one-hot encoding method. A method for searching fake news using grammar transformation on a neural network, including converting each word constituting a sentence into a vector.
3. The method of claim 1, wherein the step of generating a context vector by inputting the word vector into a predetermined natural language processing neural network comprises: generating the context vector by inputting the word vector into a long short term memory (LSTM) neural network A method of finding fake news using grammar transformation on a neural network including steps.
4. The method of claim 1, wherein generating a candidate sentence that is a sentence having the same meaning as the proposition sentence but different grammar by inputting the context vector into a predetermined natural language processing neural network comprises: sequence to sequence learning. Generating an attention vector by calculating a weighted sum of the context vector according to the model's attention mechanism; comparing the attention vector and a hidden state vector generated when the context vector is generated Predicting a matching vector; And generating the candidate sentence by inputting the matching vector into a long short term memory (LSTM) neural network.
5. The method of claim 1, Comparing the candidate sentence and the news sentence to determine the true or false of the proposition sentence and searching whether the news sentence corresponds to fake news, Corresponds to each word constituting the candidate sentence Selecting and outputting a given number of words by inputting the vector to a softmax function of a sequence-to-sequence learning model; Generating a final candidate sentence by combining the output words of the softmax function using a beam search decoder; And comparing the final candidate sentence with the news sentence to determine true or false of the proposition sentence and searching whether the news sentence corresponds to fake news.
6. The method of claim 5, Comparing the final candidate sentence and the news sentence to determine the true or false of the proposition sentence and searching whether the news sentence corresponds to fake news, The final candidate sentence and the news sentence Embedding; calculating a cosine similarity of the embedded final candidate sentence and the news sentence; And determining true or false of the proposition sentence according to the news sentence by using the cosine similarity.
7. A system for detecting fake news using grammatic transformation on neural network computer readable medium, wherein the system comprises:
a word embedding unit that embeds a proposition sentence that is determined as true or false by a news sentence as a word vector, and creates a context that generates a context vector by inputting the word vector into a predetermined natural language processing neural network;
a matching unit that inputs the context vector into a predetermined natural language processing neural network to generate a candidate sentence that has the same meaning as the propositional sentence but has a different grammar; and a reasoning unit that compares the candidate sentence with the news sentence to determine true or false of the propositional sentence and searches whether the news sentence corresponds to fake news.
8. The fake news detecting apparatus of claim 7, wherein the word embedding unit converts each word constituting the propositional sentence into a vector using a one-hot encoding method.
9. The fake news detecting apparatus of claim 7, wherein the context generation unit inputs the word vector into a LSTM (Long Short Term Memory) neural network to generate the context vector.
Application Number: Page 1 of 2 01 Dec 2020
1 2020103810
Context Word Generating Reasoning Embedding Matching Unit Unit Unit Unit 30 70 10 50
FIG. 1 is a block diagram of an apparatus for searching fake news using grammar modification on a neural network
Application Number: Page 2 of 2 01 Dec 2020
Start
Proposition Literature Embedding as a word vector v S100 2020103810
Context vector is created by inputting word vector v network into neural S200
Candidate sentences are generated by inputting v the new network context vectors into S300
Search for fake news by comparing candidate sentences withvnews sentences S400
Finish
FIG. 2 is a flowchart of a method for searching fake news using grammar modification on a neural network
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020103810A AU2020103810A4 (en) | 2020-12-01 | 2020-12-01 | A method for detecting fake news using grammatic transformation on neural network computer readable medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020103810A AU2020103810A4 (en) | 2020-12-01 | 2020-12-01 | A method for detecting fake news using grammatic transformation on neural network computer readable medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| AU2020103810A4 true AU2020103810A4 (en) | 2021-02-11 |
Family
ID=74502258
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| AU2020103810A Ceased AU2020103810A4 (en) | 2020-12-01 | 2020-12-01 | A method for detecting fake news using grammatic transformation on neural network computer readable medium |
Country Status (1)
| Country | Link |
|---|---|
| AU (1) | AU2020103810A4 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113536760A (en) * | 2021-07-06 | 2021-10-22 | 中国科学院计算技术研究所 | Rumor-rumor article matching method and system guided by introduction sentences and rumor-avoiding pattern sentences |
| CN113849599A (en) * | 2021-09-03 | 2021-12-28 | 北京中科睿鉴科技有限公司 | Joint false news detection method based on mode information and fact information |
| CN114357172A (en) * | 2022-01-07 | 2022-04-15 | 北京邮电大学 | Rumor detection method based on ERNIE-BiGRU-Attention |
| CN116596562A (en) * | 2023-07-18 | 2023-08-15 | 山东四季车网络科技有限公司 | Method and system for predicting false information risk of second-hand vehicle |
| CN118170920A (en) * | 2024-05-13 | 2024-06-11 | 南昌大学 | Mixed language text detection method and system |
-
2020
- 2020-12-01 AU AU2020103810A patent/AU2020103810A4/en not_active Ceased
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113536760A (en) * | 2021-07-06 | 2021-10-22 | 中国科学院计算技术研究所 | Rumor-rumor article matching method and system guided by introduction sentences and rumor-avoiding pattern sentences |
| CN113536760B (en) * | 2021-07-06 | 2023-09-26 | 中国科学院计算技术研究所 | Method and system for matching introduction sentence and rumor-rumor article guided by rumor-mode sentence |
| CN113849599A (en) * | 2021-09-03 | 2021-12-28 | 北京中科睿鉴科技有限公司 | Joint false news detection method based on mode information and fact information |
| CN113849599B (en) * | 2021-09-03 | 2023-01-24 | 北京中科睿鉴科技有限公司 | Joint false news detection method based on mode information and fact information |
| CN114357172A (en) * | 2022-01-07 | 2022-04-15 | 北京邮电大学 | Rumor detection method based on ERNIE-BiGRU-Attention |
| CN114357172B (en) * | 2022-01-07 | 2024-08-09 | 北京邮电大学 | ERNIE-BiGRU-attribute-based rumor detection method |
| CN116596562A (en) * | 2023-07-18 | 2023-08-15 | 山东四季车网络科技有限公司 | Method and system for predicting false information risk of second-hand vehicle |
| CN118170920A (en) * | 2024-05-13 | 2024-06-11 | 南昌大学 | Mixed language text detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020103810A4 (en) | A method for detecting fake news using grammatic transformation on neural network computer readable medium | |
| JP5128629B2 (en) | Part-of-speech tagging system, part-of-speech tagging model training apparatus and method | |
| WO2020182122A1 (en) | Text matching model generation method and device | |
| US10984785B2 (en) | Voice conversation method and system with enhanced word features | |
| JP4930379B2 (en) | Similar sentence search method, similar sentence search system, and similar sentence search program | |
| CN114254660A (en) | Multi-modal translation method and device, electronic equipment and computer-readable storage medium | |
| WO2024174911A1 (en) | Code generation method and apparatus, storage medium and electronic device | |
| CN111400504A (en) | Method and device for identifying enterprise key people | |
| US8019593B2 (en) | Method and apparatus for generating features through logical and functional operations | |
| CN113035311A (en) | Medical image report automatic generation method based on multi-mode attention mechanism | |
| CN113886601B (en) | Electronic text event extraction method, device, equipment and storage medium | |
| CN110678882A (en) | Selecting answer spans from electronic documents using machine learning | |
| CN111241850B (en) | Method and device for providing business model | |
| CN111476038A (en) | Long text generation method and device, computer equipment and storage medium | |
| KR102131641B1 (en) | Detection metohd of fake news using grammatic transformation on neural network, computer readable medium and apparatus for performing the method | |
| US20210089904A1 (en) | Learning method of neural network model for language generation and apparatus for performing the learning method | |
| CN112464655A (en) | Word vector representation method, device and medium combining Chinese characters and pinyin | |
| CN112740200A (en) | System and method for end-to-end deep reinforcement learning based on coreference resolution | |
| CN112528654A (en) | Natural language processing method and device and electronic equipment | |
| CN112364198A (en) | Cross-modal Hash retrieval method, terminal device and storage medium | |
| CN116450839A (en) | Knowledge injection and training method and system for knowledge enhancement pre-training language model | |
| CN111241843B (en) | Semantic relation inference system and method based on composite neural network | |
| Rathi et al. | Watermarking of deep recurrent neural network using adversarial examples to protect intellectual property | |
| JP2017151933A (en) | Data classifier, data classification method, and program | |
| CN112464637A (en) | Label-based optimization model training method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FGI | Letters patent sealed or granted (innovation patent) | ||
| MK22 | Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry |