Image Source: Wokandapix

Learning to Detect Fake News on Social Media

This research summary is just one of many that are distributed weekly on the AI scholar newsletter. To start receiving the weekly newsletter, sign up here.

During the 2016 US presidential election campaign, the top 20 fake news stories generated more engagement on Facebook than the top 20 major real news according to a report by BuzzFeed News. These fake news went viral on social media and earned almost 9 million shares.

To understand why MIT researchers used about 126,000 stories that had been tweeted 4.5 million times by 3 million people with data that spanned between 2006 and 2017 in a study.

Fake news has proven to travel fast and the rapid spread on social networks is alarming. This has now become a global concern because of the fact that it can negatively influence economic, political and social well being. Unlike in the past, anyone can easily start and spread false news made possible by microblogging which is a popular way for people to post, share, and seek information.

A New Dual Emotion-based Fake News Detection Framework

In this paper, researchers study the problem of learning dual emotion for fake news detection. They propose a new Dual Emotion-based fAke News detection framework (DEAN).

Two fake news posts from Chinese Weibo: (a) a post containing emotions of astonishment and sadness in news contents that easily arouse the audience, and (b) a relative objective news post without clear emotion, while emotions such as doubt and anger in user comments by controversial topics.

The framework comprises of three major components including a content module that exploits the information from the publisher, including semantic and emotion information in news contents; a comment module which captures semantic and emotion information from users; and a fake news prediction component that fuses the resultant latent representations from news content and user comments, which classifies it as fake or not.

The proposed framework DEAN consists of three components: (1) the news content module, (2) the user comments module, and (3) the fake news prediction component. The previous two modules are used to model semantic and emotions from the publisher and users respectively, while the prediction part fuses information from these two modules and makes a prediction.

The framework can learn content- and comment- emotion representations for publishers and users respectively. DEAN is also able to exploit the dual emotion representations simultaneously for fake news detection.

Why Detecting Fake News is Important?

The generation and propagation of fake news results in detrimental societal consequences.

Now more than ever before, it is important to intensify research efforts that will build tools that detect fake news automatically and effectively.

The proposed DEAN framework can capture and integrate the dual emotion jointly for learning news representation to detect fake news. On evaluation with Weibo and Twitter datasets, DEAN demonstrates its effectiveness by outperforming several state-of-art fake news detection methods.

Read more: Fake News Detection on Social Media

Thanks for reading, comment, share & let’s connect on Twitter, LinkedIn, and Facebook. Stay updated with the latest AI research developments, news, resources, tools, and more by subscribing to our weekly AI Scholar Newsletter for free! Subscribe here Remember to 👏 if you enjoyed this article. Cheers!