WO2021256884A1

WO2021256884A1 - Method and device for identifying fake news using ai

Info

Publication number: WO2021256884A1
Application number: PCT/KR2021/007642
Authority: WO
Inventors: 김성수; 이주훈
Original assignee: 주식회사 푸시뉴스
Priority date: 2020-06-17
Filing date: 2021-06-17
Publication date: 2021-12-23
Also published as: KR102228737B9; KR102228737B1

Abstract

According one aspect of the present embodiment, provided is a fake news identification device characterized by comprising: a news information collection unit which collects stock news information from news providers; a news sorting unit which filters out, from the stock news, only item-of-interest news information corresponding to items of interest or keywords set in a subscriber's terminal; a fake news determination unit which generates an authenticity determination result obtained by using artificial intelligence to determine whether the item-of-interest news information is fake news on the basis of learned training data; and a news curation unit which matches the authenticity determination result with the item-of-interest news information to curate the stock news for the subscriber's terminal.

Description

Method and device for reading fake news using AI

One embodiment of the present invention relates to a method and apparatus for reading fake news using AI.

The content described below merely provides background information related to the present embodiment and does not constitute the prior art.

The news curation market in the securities sector is estimated to be approximately 1.5 trillion won as of 2019. It is estimated that the main consumers are 5.6 million to 8 million individual investors. Due to the recently passed Data 3 Act, it is expected that marketing and business will be possible even with non-real name information.

There are approximately 5.6 to 8 million individual investors who trade stocks while looking at domestic stock information, but they are indiscriminately exposed to fake news. In general, individual investors use the Internet to search for relevant information, but they waste time and money on multiple sites, and cannot find the desired and accurate information.

Recently, the consumption environment is changing due to the increase in single-person media. Since single-person media news can be consumed in real time regardless of time and place, the side effect of producing and distributing reckless fake news such as interesting creations or combinations is increasing.

The economic and social harm of fake news is increasing. In other words, 'fishing articles', 'abusing articles' for the purpose of inducing traffic, and 'advertising articles' written for a specific purpose such as publicity are aggravating the confusion of consumers who consume news. Social costs are also increasing significantly, and the economic damage caused by fake news is estimated to reach 30 trillion won.

National credibility due to fake news is declining, and according to a Reuters Media Research Institute survey, South Korea is ranked as the worst fake news country in the world with 23% of news credibility.

Due to the information asymmetry problem, the damage caused by fake news in the reliability of information is increasing rapidly. Therefore, issues of reliability and fairness are emerging, and technologies that contribute to resolving social, economic, and political conflicts by clearly discriminating news content and the like are required.

This embodiment collects news and provides reliable information according to consumer preferences when providing a service that curates and shows various news in the securities information market to individual investors who are fragmented from various websites to find securities information, , aims to provide a fake news reading method and device using AI that uses artificial intelligence to read fake news and distinguish fake news and reliable news from among securities information.

According to an aspect of this embodiment, a news information collection unit for collecting stock news information from a news provider; a news selection unit filtering only the news information of the items of interest corresponding to the keywords or items of interest set in the subscriber terminal among the stock news information; a fake news determination unit for generating a truth determination result for determining whether the news information on the item of interest is fake news or real news based on previously learned learning data using artificial intelligence; and a news curation unit that matches the authenticity determination result to the interest item news information and curates the news to the subscriber terminal.

According to another aspect of this embodiment, the process of collecting stock news information from a news provider in the news information collection unit; a process of filtering, in the news selection unit, only the news information of the interest item corresponding to the keyword or the item of interest set in the subscriber terminal among the securities news information; a process of generating a truth determination result for determining whether the news information of the item of interest is fake news or real news based on previously learned learning data using artificial intelligence in the fake news determination unit; and a news curation unit curating the news information on the item of interest and the authenticity determination result to the subscriber terminal.

As described above, according to this embodiment, when providing a service that curates and shows various news in the securities information market to individual investors who are fragmented from various websites and looking for securities information, news is collected and reliable information is provided according to consumer preferences, and by using artificial intelligence to read fake news, it has the effect of discriminating between fake news and reliable news among securities information.

According to the present embodiment, there is an effect that the authenticity information can be determined by automatically updating the authenticity by performing the authenticity determination process according to the reliability of the expert or the fact check of the expert in each field as the basis for determining the fake news.

1 is a diagram schematically illustrating a fake news reading system using AI according to the present embodiment.

2 and 3 are diagrams schematically showing a fake news reader according to the present embodiment.

4 is a diagram illustrating a process of providing a fake news reading service according to the present embodiment.

5 is a diagram illustrating a method of providing fake news identification information according to the present embodiment.

6 is a diagram illustrating a context determination method for reading fake news according to the present embodiment.

7 is a diagram illustrating a method of determining whether to synthesize for fake news reading according to the present embodiment.

8 is a diagram illustrating a learning execution concept of a fake news reader according to the present embodiment.

9 is a diagram illustrating a news authenticity determination service using artificial intelligence according to the present embodiment.

10 is a diagram illustrating a service provision concept according to the present embodiment.

Hereinafter, this embodiment will be described in detail with reference to the accompanying drawings.

The fake news reading system using AI according to the present embodiment includes a subscriber terminal 110 , a fake news reader 120 , a manager terminal 130 , a news provider 140 , and an expert judgment group terminal 150 . Components included in the fake news detection system using AI are not necessarily limited thereto.

The subscriber terminal 110 receives news content and news authenticity information from the fake news reader 120 . The subscriber terminal 110 provides points and rewards according to the subscription from the fake news reader 120 .

The subscriber terminal 110 accesses the fake news reader 120 to search for news content and receives news subscription information.

The subscriber terminal 110 requests the authenticity of the news through the fake news reader 120 and uses a paid service.

The subscriber terminal 110 manages membership registration and withdrawal. The subscriber terminal 110 manages member information, charges points, and manages the use of points. The subscriber terminal 110 applies for the authenticity determination service and confirms news content information. The subscriber terminal 110 files an objection to the authenticity of the news content.

The fake news reader 120 curates stock information and provides it as a mobile application. The fake news reader 120 provides news curation that filters out fake news using big data on stock information and artificial intelligence.

The fake news reader 120 collects securities-related data by using RSS and crawling technology from the news provider 140, which is the first source of transmission of securities news, by contracting with the transmission medium of the securities news.

The fake news reader 120 classifies confirmed and estimated articles on the collected securities-related data, and determines the authenticity of the news through a process of merging big data. The fake news reader 120 analyzes, using an algorithm, whether stock price changes, news, and social media are transmitted in the securities field after transmission.

The fake news reader 120 informs the probability that the content of the news source specified by the user is fake using artificial intelligence. The fake news reader 120 provides a service for determining the authenticity of news using artificial intelligence based on the fact check result received from the expert judgment group terminal 150 by field experts and verified judges for accurate fact checking. 110) is provided.

The fake news reader 120 uses artificial intelligence technology to sort out fake news and reliable information from among the securities information desired by the user. The fake news reader 120 provides a basis for discriminating between fake news and reliable information. The fake news reader 120 determines the authenticity information by automatically updating the authenticity according to the judgment information received from the expert judgment group terminal 150 participating in the expert's reliability or authenticity with respect to the evidence.

The fake news reader 120 provides, to the subscriber terminal 110 , important information on stock information and trend information of companies, which are important to companies, organizations, and individual investors who need stock information. The fake news reader 120 learns about the results for determining the authenticity of news content based on artificial intelligence.

The manager terminal 130 accesses the fake news reader 120 and controls overall functions of the fake news reader 120 . The manager terminal 130 manages the authenticity of the content.

The manager terminal 130 manages member and expert information. The manager terminal 130 manages approval and suspension of experts. The manager terminal 130 manages the point deposit corresponding to the real or fake determination information. The manager terminal 130 manages content authenticity information. The manager terminal 130 manages the expert profit policy.

The news provider 140 receives a news search (keyword, subscription) request from the fake news reader 120 , and transmits news information corresponding to the news search to the fake news reader 120 .

The expert judgment group terminal 150 transmits real or fake judgment information corresponding to the news information to the fake news reader 120 . The expert judgment group terminal 150 is provided with points or rewards corresponding to real or fake judgment information from the fake news reader 120 .

The expert judgment group terminal 150 is a registered expert or judgment group terminal, and sets a field of expertise. The expert judgment group terminal 150 checks and manages the authenticity of stock news information corresponding to a preset specialized field.

The expert judgment group terminal 150 receives points corresponding to real or fake judgment information and manages the accumulated status and usage history. The expert judgment group terminal 150 applies for and manages objections to a preset specialized field. The expert judgment group terminal 150 receives points corresponding to the genuine judgment information or the fake judgment information.

The fake news reader 120 includes a fake news learning engine using artificial intelligence technology and a fake news determination engine using artificial intelligence technology. The fake news reader 120 manages learning data, manages subscription information settings, calculates a fake news probability, and provides information.

The fake news reader 120 according to this embodiment includes a news information collection unit 210 , a news selection unit 220 , a fake news determination unit 230 , a learning unit 232 , a truth determination unit 234 , and a news queue. It includes a ration unit 240 . Components included in the fake news reader 120 are not necessarily limited thereto.

In other words, the fake news reader 120 is a member management module, expert management module, point management module, subscription information management module, payment management module, service purchase management module, authenticity probability management module, news content information management module, expert evaluation module , it can be implemented in the form of including an opinion management module for authenticity evaluation by content, a setting management module for specialized fields by experts, a fake news identification policy management module, an expert ranking management module, a truth determination module, a notification notification module, and an objection module.

Each component included in the fake news reader 120 may be connected to a communication path that connects a software module or a hardware module inside the device to operate organically with each other. These components communicate using one or more communication buses or signal lines.

Each component of the fake news reader 120 shown in FIGS. 2 and 3 means a unit that processes at least one function or operation, and may be implemented as a software module, a hardware module, or a combination of software and hardware. have.

The news information collection unit 210 collects stock news information from the news provider 140 . The news selection unit 220 filters only the news information of the items of interest corresponding to the keywords or items of interest set in the subscriber terminal 110 among the securities news information.

The fake news determining unit 230 generates a result of determining whether the news information of the item of interest is fake news or real news based on previously learned learning data using artificial intelligence. The fake news determining unit 230 determines whether the news information of the interest item is fake news based on the expert judgment information (the expert real judgment information or the expert fake judgment information) and learning data.

The fake news determining unit 230 separates the text and image included in the news item of interest information and performs pre-processing on the text. The fake news determining unit 230 recognizes a word based on the characters. The fake news determining unit 230 understands the context based on the order of combining words, and generates its own fact check result information based on the context.

The fake news determining unit 230 stores the contract amount, the contract year, and the contract conditions as fact information if it is determined that the contract or the order article of a specific company is based on the context. The fake news determining unit 230 stores the investment destination, investment field, investment cost, and investment year as fact information when determining that it is an investment article of a company or government based on the context. The fake news determining unit 230 stores the increase/decrease rate and date and time of the stock market by country as fact information when it is determined as a world stock market article based on the context. The fake news determining unit 230 stores the new technology information for each company as fact information if it is determined that the new technology development article for each company is based on the context. The fake news determining unit 230 stores the economic growth rate information for each country as fact information when it is determined that the article is an article on economic growth rates for each country based on the context.

When new stock information is recognized based on the context, the fake news determining unit 230 checks whether the new stock information and pre-stored fact information match, and generates its own fact check result information. As a result of the check, if the new stock information matches the previously stored fact information, the fake news determining unit 230 generates its own fact check result information determined as real news. As a result of the check, if the new stock information does not match the previously stored fact information, the fake news determining unit 230 generates fact check result information determined as fake news.

The fake news determining unit 230 separates the text and the image included in the news item of interest information and generates image determination result information confirming whether the image is synthesized or not. As a result of the check, if it is determined that the image is synthesized, the fake news determining unit 230 generates fake image determination information. As a result of the check, if it is determined that the image is not synthesized, the fake news determining unit 230 generates image authenticity determination information.

The fake news determining unit 230 calculates fake news probability information when it is determined that the news information of the item of interest is fake news. The fake news determining unit 230 includes expert judgment information (expert genuine judgment information or expert fake judgment information), self-fact check result information (real news, fake news), image judgment result information (image real judgment information or image fake judgment information) ) to calculate fake news probability information.

The learning unit 232 analyzes fake news cases and fake news patterns. The learning unit 232 learns fake news cases and fake news patterns and reflects them in the learning data. The learning unit 232 learns the result of analyzing stock news information based on fake news cases and fake news patterns and reflects them in the learning data.

When new stock news information is received from the news provider 140 , the authenticity determining unit 234 classifies a field of the new stock news information before delivering it to the subscriber terminal 110 .

The authenticity determination unit 234 transmits the stock news information to the expert determination group terminal 150 having a field of expertise corresponding to the field. The authenticity determining unit 234 receives expert judgment information (real expert judgment information or expert fake judgment information) on new securities news information according to the field of expertise from the expert judgment group terminal 150 .

The news curation unit 240 performs curation with the subscriber terminal 110 by matching the authenticity determination result with the news information of the item of interest. The news curation unit 240 transmits the fake news probability together with the news information of the item of interest.

As shown in (a) of FIG. 4 , the subscriber terminal 110 sets up a newspool that the subscriber is interested in. As shown in (b) of FIG. 4 , the subscriber terminal 110 sets keywords and outputs interesting news information as a search result.

As shown in (c) of FIG. 4 , the subscriber terminal 110 retrieves information corresponding to a search term from the news provider set by the user from the fake news reader 120 and outputs it to the news list window.

As shown in (d) of Fig. 4, when the user clicks on news information of interest, the subscriber terminal 110 determines the authenticity information analyzed from the fake news reader 120 and the authenticity presented by the expert in the detailed information window. Information that can be checked is placed at the top and output, and the corresponding content information is provided along with advertisements.

First, the fake news reader 120 determines whether the news information is fake news or real news using expert judgment information (real expert judgment information or expert fake judgment information) received from the expert judgment group terminal 150 . That is, the fake news reader 120 performs a professional judgment step to determine the authenticity of the news content.

Secondary, the fake news reader 120 recognizes the text inserted in the news, analyzes the context, and compares it with pre-stored fact information to determine the authenticity of the news by its own fact check. That is, the fake news reader 120 determines the probability of fake news by grasping the context of the news content.

Thirdly, the fake news reader 120 determines the authenticity of the news by determining whether the image inserted into the news is a composite photo.

That is, the fake news reader 120 determines whether the photo information included in the news is synthesized to determine the probability of fake news.

The fake news reader 120 receives the expert real judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is real news by its own fact check, and the image inserted into the stock news information is an image If it is judged by the real judgment information, the corresponding news is judged as real news.

The fake news reader 120 receives the expert fake judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is real news by checking its own facts, and the image inserted into the stock news information is an image If it is determined as genuine information, it indicates that the news may be fake news as a result of the expert's judgment on the news, and the probability information of fake news is calculated and output together with the news.

The fake news reader 120 receives the expert real judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is fake news, and determines that the text included in the stock news information is fake news, and the image inserted in the stock news information is an image. If it is judged as genuine information, it indicates that the news may be fake news as a result of the self-fact check, and the probability information that the self-pack check result is fake news is calculated and output together with the news.

The fake news reader 120 receives the expert real judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is real news by its own fact check, and the image inserted into the stock news information is an image If it is judged as fake information, the image pack check for the news indicates that it may be fake news, and the probability information of fake news including fake images (synthetic images) is calculated and output together with the news.

Here, since the fake news reader 120 determines that one of the three criteria is fake (expert judgment information, self-fact check result information, image judgment result information), the fake news reader 120 calculates, for example, about 30% of the probability information of fake news. can do.

When it is determined that the expert judgment information is fake news, the fake news reader 120 may calculate fake news probability information according to the expert field determined by the expert. The fake news reader 120 reflects a weight higher than the average threshold when the specialized field is a quantifiable field (eg, IT field), and is higher than the average threshold when the specialized field is a non-quantifiable field (eg, politics, current affairs, prediction). By reflecting the low weight, fake news probability information can be calculated.

When the fake news reader 120 determines that the fact check result information is fake news, the fake news reader 120 can calculate fake news probability information by counting the number of mismatches with the items included in the pre-stored fact information. The fake news reader 120 may calculate fake news probability information by reflecting a weight according to importance to each item in the pre-stored fact information.

The fake news reader 120 calculates fake news probability information in connection with the context included in the news or quantifies the synthesis or forgery probability when it is determined by the image determination result information and it is confirmed that the image is synthesized or forged image inserted in the news. Thus, fake news probability information can be calculated.

The fake news reader 120 calculates fake news probability information based on a result determined as fake news among news determination results, expert determination information, own fact check result information, and image determination result information.

For example, the fake news reader 120 receives the expert fake judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is fake news by checking its own facts, and the image inserted into the stock news information If is judged by the image real judgment information, it indicates that the expert judgment result and the self fact check result for the news may be fake news. It should be printed along with the news.

In other words, the fake news reader 120 determines that two of the three judgment criteria are fake (eg, expert judgment information (expert fake judgment information), fact check result information (fake news), image judgment result information (image real) decision information)), so the probability information of fake news can be highly calculated (eg, about 60%).

For example, the fake news reader 120 receives the expert fake judgment information from the expert judgment group terminal 150, determines that the text included in the stock news information is fake news by checking its own facts, and the image inserted into the stock news information If is judged as image fake judgment information, it indicates that the expert judged the news, the self-packed check result, and the image judgment result may all be fake news, and the expert judgment result, the self-packed check result, and the image judgment result are all fake. The probability information of news is calculated and output together with the corresponding news.

Here, the fake news reader 120 judges all three judgment criteria to be fake (expert judgment information (expert fake judgment information), fact check result information (fake news), image judgment result information (image fake judgment information)), so it is fake. It is possible to calculate the probability information of news highly (eg, about 90%).

According to an embodiment of the present invention, the determination of the text for identifying fake news may be performed based on the following method.

A determination of fake news may be performed based on the learning result of the association between words included in the text. Learning about existing real news may be performed for news discrimination. After learning on the degree of relevance between words through learning about real news, the judgment on fake news is performed by additionally considering the judgment on word relevance included in the fake news to determine the authenticity of fake news. can Learning about word relevance included in real news can be performed by extracting primary and secondary relevance between words. The primary and secondary associations may be extracted by considering the distance between words in the text. The distance between words for determining the primary and secondary associations may be adaptively adjusted according to the length of input news for input learning. A degree of relevance between words is extracted based on the first degree of relevance and the second degree of relevance, and a relatively high weight is placed on the first degree of relevance to determine the authenticity of the news.

In addition, according to an embodiment of the present invention, in order to determine the authenticity of the news, information on the publisher of the news, information on the publication speed of the news, and the information on the amount of publication may be additionally learned. For the same topic, information on the publisher, the amount of issuance, and the speed at which existing real news and fake news are generated as news can be collected, and the probability that the published news is fake news is It can be adjusted according to the flow.

For example, in the case of real news, similar news from different sources may have a tendency to increase and then gradually decrease based on the publication time of one news. However, in the case of fake news, it is rapidly published from the same source, and there is no increase in news from other sources, and it can disappear rapidly. Through learning about these news sources, news issuance volume, and news publication speed, learning about the publication patterns of fake news and real news spreading can be performed, and based on this, a judgment as to whether news is real news or fake news can be made in real time. It can be done with change.

The fake news reader 120 uses Recurrent Neural Networks (RNNs) to determine a context from data containing time series information, such as a character string and video, and identifies fake news based on the context.

The fake news reader 120 has an artificial neural network structure composed of a plurality of layers, and the fake news reader 120 pre-processes sentences in stock news information and extracts features. The fake news reader 120 recognizes the words included in the stock news information, understands the context based on the order of the words, and remembers when it is determined that the words appearing in the context are important words for understanding the context.

When the fake news reader 120 determines the context and determines that the corresponding securities information is fact-checked information, it is stored. For example, when the fake news reader 120 determines that the contract or order information article of a specific company is a fact, the contract amount, the contract year, and the contract conditions are stored. Thereafter, when a new article is found in the same field, the fake news reader 120 may search whether the new article is fake news based on the fact-checked information. The fake news reader 120 has a high learning effect when the difference (GAP) between the pre-stored information and the place where the information is needed is small.

The fake news reader 120 identifies fake news by determining whether a photo included in the stock news information is synthesized. The fake news reader 120 uses a convolutional neural network (CNN) to extract and learn features by using convolution and pooling.

The fake news reader 120 uses the output data of the convolution again as input data. The fake news reader 120 is used to reduce or emphasize data. When the fake news reader 120 performs pooling, Max Pooling, Min Pooling, and Average Pooling may be used. However, CNN mainly uses maximum pooling.

The fake news reader 120 performs learning to read fake news. The fake news reader 120 analyzes fake news cases and patterns, and learns a result of analyzing stock news information based on fake news cases and patterns.

The fake news reader 120 performs learning using a Recurrent Neural Networks (RNN) algorithm suitable for handling data containing time series information, such as character strings and moving pictures, for stock news information composed only of text without images.

The fake news reader 120 may learn the part related to image synthesis using a Convolutional Neural Network (CNN) of TensorFlow.

The fake news reader 120 automatically determines the probability of fake news by analyzing news information based on the learned data, starting from supervised learning using learning data to determine the authenticity of the news using artificial intelligence technology. It is provided to users through the news authenticity service.

As shown in FIG. 9 , the news content and creator information input module transmits the received stock news information to the fake news content information operation module and the fake news content creator information processing module for learning data processing.

The fake news content information operation module extracts news content from the received stock news information. The fake news content creator information processing module extracts author information from the received stock news information.

The authenticity information verification module verifies news information authenticity information for news content. The news authenticity determination module checks whether the news is fake news by performing a fact check using the news learning information on the news content. The author propensity determination module classifies the author propensity by using the author propensity learning information based on the author information.

The news classification module classifies news based on fake news information and author tendency in news content. The news authenticity information output module outputs news authenticity information. The news information input module transmits news authenticity information to the web server and the communication server.

The fake news reader 120 receives information from a source source of news content, such as a news medium. In other words, the fake news reader 120 collects stock news information from the media news provider 140 . The fake news reader 120 extracts only news information in a specific field corresponding to a keyword or field set in the subscriber terminal 110 from among the collected stock news information.

The fake news reader 120 analyzes the probability that news information in a specific field is fake news using artificial intelligence. The fake news reader 120 provides the analyzed probability information of fake news together with news information in a specific field to the user. The subscriber terminal 110 may set up a subscription for a news medium and separately check related information by filtering items of interest through keywords.

The above description is merely illustrative of the technical idea of this embodiment, and a person skilled in the art to which this embodiment belongs may make various modifications and variations without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain rather than limit the technical spirit of the present embodiment, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be interpreted as being included in the scope of the present embodiment.

Claims

a news information collection unit that collects stock news information from news providers;

a news selection unit filtering only the news information of the items of interest corresponding to the keywords or items of interest set in the subscriber terminal among the stock news information;

a fake news determination unit for generating a truth determination result for determining whether the news information on the item of interest is fake news or real news based on previously learned learning data using artificial intelligence; and

and a news curation unit that matches the authenticity determination result with the interest item news information to curate the subscriber terminal, wherein the fake news determination unit separates the text and image included in the interest item news information. After performing pre-processing on the characters, words are recognized based on the characters, and contexts are recognized based on the order in which the words are combined,

If it is determined as a contract or order article of a specific company based on the context of the news information on the item of interest, the contract amount, contract year, and contract condition are saved as fact information The investment destination, investment field, investment cost, and investment year are stored as fact information. Based on the above context, if it is judged as a world stock market article, the change rate and date and time of the stock market by country are stored as fact information, and new technology development for each company based on the above context If it is judged as an article, new technology information for each company is stored as fact information, and if it is judged as an article on economic growth by country based on the above context, information on economic growth rate for each country is stored as fact information, and new securities information is recognized based on the context. When the new stock information matches the pre-stored fact information, the fact check result information determined as real news is generated, and when the new stock information does not match the pre-stored fact information, the fact check result information determined as fake news create,

adaptively according to the sentence length in the fact information, extracting the primary and secondary associations based on the distance between words included in the sentences in the fact information, and based on the primary and secondary associations After extracting the degree of relevance between the words as , it is determined whether the news information of the item of interest is fake news or real news by reflecting a higher weight than the second degree of relevance in the first degree of relevance,

After collecting the publisher information, the publication rate information, and the publication amount information for the news item of interest, it is confirmed that similar news is issued from another publisher based on one publisher based on the publication information, and the issuance rate information and the Based on the publication amount information, if the news is issued in an increasing trend and then gradually decreases, the fact check result information determined as the real news is generated, and it is confirmed that similar news is issued only in the same publisher based on the publication information and, based on the issuance rate information and the issuance amount information, generates fact check result information determined as fake news when news is issued rapidly and then rapidly disappears,

It is checked whether the image is synthesized, and if it is determined that the image is synthesized, image fake determination information is generated. Receive expert real judgment information and expert fake judgment information for the new securities information accordingly;

The image real determination information, the image fake determination information, the expert genuine determination information, the expert fake determination information, the fact check result information determined as the real news, and the fact check result information determined as the fake news. Calculate fake news probability information based on the ratio,

When calculating the fake news probability information,

For the real expert judgment information and the expert fake judgment information, if the specialized field is a quantifiable field, a weight higher than the average threshold is reflected, and if the specialized field is a quantifiable field, a weight lower than the average threshold is reflected,

Counting the number of items inconsistent with the items included in the fact information stored in the fact check result information determined as fake news,

A fake news reading apparatus, characterized in that the weight according to the importance of each item in the pre-stored fact information is reflected, and the synthetic or forgery probability is digitized in the image fake determination information and reflected in the fake news probability information.
The method of claim 1,

Fake news cases and fake news patterns are analyzed, the fake news cases and the fake news patterns are learned and reflected in the learning data, and the stock news information is analyzed based on the fake news cases and the fake news patterns. Fake news reading device, characterized in that it further comprises a learning unit for learning and reflecting in the learning data.
3. The method of claim 2,

When the new stock information is received from the news provider, the new stock information field is classified before being delivered to the subscriber terminal, and the stock news information is transmitted to the expert judging group terminal having a specialized field corresponding to the relevant field. and an authenticity determining unit for receiving expert judgment information on the new securities information according to the specialized field from the expert judgment group terminal.
4. The method of claim 3,

The fake news discrimination unit,

and determining whether the news information on the item of interest is fake news or not based on the expert determination information and the learning data.
a process of collecting stock news information from news providers in the news information collection unit;

a process of filtering, in the news selection unit, only the news information of the interest item corresponding to the keyword or the item of interest set in the subscriber terminal among the securities news information;

After the fake news determination unit separates the text and the image included in the news item of interest information, pre-processing is performed on the text, the word is recognized based on the text, and the context is determined based on the order of combining the words. process of perception;

When the fake news determination unit determines that the contract or order for a specific company is a contract or order article based on the context of the news information on the item of interest, the contract amount, contract year, and contract conditions are stored as fact information, and based on the context, the When it is judged as an investment article of a process of storing new technology information for each company as fact information if it is determined as a new technology development article for each company based on the basis, and storing the economic growth rate information for each country as fact information if it is determined as an article on economic growth rate by country based on the context;

When the fake news determining unit recognizes new stock information based on the context, if the new stock information matches pre-stored fact information, it generates fact check result information determined as real news, and the new stock information is pre-stored. The process of generating fact check result information determined to be fake news when the fact information does not match;

The fake news determining unit extracts the primary and secondary associations based on the distance between words included in the sentences in the fact information adaptively according to the sentence length in the fact information, and the primary association and the After extracting the degree of relevance between words based on the second degree of relevance, a weight higher than the second degree of relevance is reflected in the first degree of relevance to determine whether the news information of the item of interest is fake news or real news. the process of creating;

collecting, in the fake news determining unit, information on a publisher of the news item of interest, information on a publication rate, and information on an amount of publication;

In the fake news determination unit, it is confirmed that similar news is issued from another publisher based on one publisher based on the information on the publisher based on the information on the publisher. generating fact check result information determined as the real news when it has a decreasing trend;

When the fake news determination unit confirms that similar news is issued only from the same publisher based on the information on the publisher, and the news is issued rapidly and then rapidly disappears based on the information on the issuance rate and the amount of issuance, the The process of generating fact check result information determined to be fake news;

checking whether the image is synthesized by the fake news determining unit, generating image fake determination information if it is determined that the image is synthesized, and generating image genuine determination information if it is determined that the image is not synthesized;

receiving, in the fake news determination unit, real expert judgment information and expert fake judgment information for the new securities information according to the field of expertise from the expert judgment group terminal;

Among the fake news determination unit, the image real determination information, the image fake determination information, the expert real determination information, the expert fake determination information, the fact check result information determined as the real news, and the fact check result information determined as the fake news A process of calculating fake news probability information based on a ratio including fake determination information;

In the fake news determination unit, a weight higher than the average threshold value is reflected when the specialized field is a quantifiable field for the expert real determination information and the expert fake determination information, and a weight lower than the average threshold value is applied when the specialized field is a field that cannot be quantified. applying and reflecting the fake news probability information;

In the fact check result information determined by the fake news determination unit as fake news, the number of items inconsistent with the items included in the stored fact information are counted, and a weight is applied according to the importance of each item in the stored fact information to apply the fake news. The process of reflecting the probability information;

a process of digitizing the probability of synthesis or forgery in the fake image determination information by the fake news determination unit and reflecting it in the fake news probability information; and

and curating, by a news curator, the news item of interest, the authenticity determination result, and the fake news probability information to the subscriber terminal.