1、 Overview

At present, the main platforms for identifying and refuting rumors in China basically rely on expert identification mode, which has certain problems: the discovery of clues mainly depends on user reports, with limited number and weak timeliness; The speed of news authentication needs to be improved. In order to improve the efficiency of rumor recognition, the People's Daily Online, in cooperation with the Institute of Computing of the Chinese Academy of Sciences, has adopted "artificial intelligence" technology as an effective auxiliary tool to establish an efficient rumor recognition platform - news credibility recognition support system.

2、 Implementation mode

The news credibility identification support system jointly built by People's Daily Online and the Institute of Computing of the Chinese Academy of Sciences makes full use of artificial intelligence, knowledge maps, and all media credible big data training. With knowledge content as the basic framework, it comprehensively analyzes, deduces, and summarizes the precise needs of communication content cognition, automatically and timely finds suspicious clues and authenticates them, greatly reducing the potential harm of rumors; With the aid of machine learning algorithm and manual review, it only takes 1 minute to give an early warning for suspected rumor events; Based on the data-driven method, the platform can also continuously mine the characteristics of different types of rumors, and realize the automatic identification of various rumors.

（1） Core technology

In terms of core technology, AI relies on "three more" to screen rumors:

One is multimodal data. Rumors may be accompanied by text, pictures, videos, communication networks, participating user attributes and other multimodal data in the life cycle from publishing and spreading to being refuted. All modal data can indicate rumors to varying degrees, such as the emotional tendency of rumor words, the visual impact of pictures, and the structural attributes of communication networks.

The second is multi-level representation. Deep learning technology has strong representation learning ability. Through the design of neural network structure and fusion mechanism, When facing numerous modal data with different meanings, forms, and structures, the AI model can integrate data at different levels, automatically find the most discriminating representation combination, and fuse and project the text, pictures, and communication networks corresponding to the articles to be screened into the feature space.

Third, judge from multiple perspectives, AI can give credibility from a single aspect (such as content, users, and communication), or it can comprehensively observe, give credibility of all information, and timely help people "find" information with low credibility.

（2） Calculation method

First, the model will trace information sources and analyze user profiles. It includes: professional field, previously disseminated or forwarded content, subject nature, registration time, active rule, etc. to judge whether the publisher is "reliable". Finally, the credibility of the news is calculated according to the proportion of people with different attitudes, their credibility and other information.

Second, confirm whether the source of information is credible. The system will search all information sources on the Internet to see if the linked domain name is from a trusted website. For example, whether it comes from official websites such as Xinhua News Agency and the government medical administration.

Third, refining and verification. The key arguments in the text are refined into knowledge points and matched with the authoritative knowledge base in the knowledge map for verification. If there is no connection or contradiction, it will be automatically identified.

3、 Application module

（1） Content production module

It supports the discovery of controversial news clues on the Internet within the fastest 10 minutes, timely warning of high-risk suspicious events, and in-depth three-dimensional analysis of events, showing the key participants, key regions, relevant news, event development context and other multidimensional evidence chains of events, and assisting manual review. At the same time, the identification system of illegal content can BK、 Filter content from the perspective of political sensitivity and video copyright.

（2） De fake and truth seeking module

In addition to encouraging netizens to actively disclose information through microblog and WeChat, the People's Daily Online's "Seeking Truth" column also ensures to obtain investigation clues at the first time through the ability to identify rumors provided by the intelligent rumor dispelling platform. The close cooperation of netizens, media and platforms has greatly expanded the channels of information sources, and then formed an efficient and professional channel to refute rumors.

（3） Public service module

Epidemic topic - 24 hour controversial news broadcast system of COVID-19 epidemic: support the discovery of controversial clues in 10 minutes at the earliest; Support searching by hot topics; It supports retrieval by geographical classification (epidemic map). Provide timely and accurate thematic information for the public.

4、 Application effect

By embedding the ability of the credibility recognition support system into the news monitoring and content management system, the People's Daily Online allows AI to participate in the content production process through the means of automatic news review and editing of credibility information, which basically covers the news content production process of all channels of the People's Daily Online, greatly improving the accuracy of news and the efficiency of content production. At present, this system has been included in the "2020 Chinese Newspaper Industry Deep Integration Development Innovation Case" of the selected Propaganda Department.

5、 Conclusion

With the continuous progress and development of Internet technology, the intelligent rumor recognition platform will further enhance the technical capability of "artificial intelligence+big data", actively expand the research on the direction of "the application of intelligent computing facilities in the field of content dissemination", and build a more efficient, more accurate and broader media "anti fraud platform" in combination with the massive media resource database of People's Daily Online, It provides the industry with a "false information database", high-quality data support and application scenarios for the National Key Laboratory of Communication Content Cognition, escort more platforms, provide more people with true and accurate media information, and make continuous contributions to building a safe and clean information environment for society.

Share to let more people see