Social media remains to be one of the most debated sources of information. While it provides for cheap, available-to-all, and rapid news sources allowing its users to react to news in an unfiltered, timely manner, “fake news,” or news with no or nebulous credibility finds a way into circulation. False information may be intentional (for instance, to polarize users towards a particular organization or school of thought) or unintentional (for example, information not fact-checked).
Users are both creators and consumers of news on social media. Thus, it becomes a shared responsibility of content creators, consumers, and the social media platform to ensure a safe and trusted space for all.
When encountering a post, naive realism (the belief that perceived objects are not representations of said objects, rather the objects themselves) and confirmation bias (the tendency to accept information that confirms existing beliefs) kick in. These collectively cause the users to side with the source and trigger the “gossip model.”
The Gossip Model
An easy way to liken the concept of fake news propagation is through the “gossip” analogy, with some modular changes. Also, note that every user adds personal flavor or response to the news.
Malicious fake news is often decorated with attention-grabbing titles, compelling pictures, and clickbait to lure users into sharing. The effect of the process described in the figure is mass polarization that can even lead to actual (and often unfavorable) action.
Current Research in the Domain
With the current trends in artificial intelligence, fake news, with its myriad characteristics, has been greatly simplified for ease of detection. Further, with deep learning, models today can recognize complex patterns in the underlying data fed to them.
The problem of fake news detection can be stated as follows:
Given a social media port P, with features F, find out whether the information I contained in P is fake.
Features are simply inputs to the machine learning model.
Think about what the features of a news article could be:
- the source (authoring website or blog),
- the headline (a top-level attribute of the news article),
- body text (a low-level attribute of the news article),
- images or videos attached.
We can dive deeper into these characters.
For the source, past records of the publisher (in the case of social media, the account) can help track the problem better. Headlines and body can be decomposed using Natural Language Processing into their constituents, such as sentiment (negative, positive, or neutral), language style (writing style, deceptive statements, vagueness, subjectivity), and bag-of-words (words and their frequencies).
Apart from these, features from the social media platform themselves can be leveraged. Most intentional fake news posts use social media bots to generate fake likes, comments, or shares. The credibility of users reacting to the post says a lot about the post. Therefore, an analysis of the type of engagement on a post is another useful feature.
User credibility, social media bot detection, and post propagation are established research problems in themselves.
Traditional machine learning models such as Support Vector Machines (SVMs) and Random Forests can be applied on top of these features. These models are trained (which is to say, their variable weights are fixed) through available datasets.
These days, deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) such as LSTMs have been popularized. An exciting trend in research is the focus on explainable deep learning models. This is because deep learning models are often treated as a “black box,” which spit out a binary result (in this case, fake or not fake) with some confidence percentage. Explainable models allow for a deconstruction of the result to gain more insights into the “why” of a particular result. An example is the Event Adversarial Neural Network (EANN) proposed by Wang et al. in 2018.
There are publicly available annotated datasets such as BuzzFeed News, LIAR, and CREDBANK that researchers test their detection models on.
To mitigate the effect of fake news on social media, browser extensions, and add-ons such as the Fake News Guard, FakerFact and others, and Google’s Fact Check Explorer that use detection models can be utilized. While these tools are helpful, it is crucial to understand that this is a community responsibility. Be conscious while sharing news online. Similarly, do not be shy to admit a previously shared fake post and educate your followers.
A simple way to state this would be, every time you share a post or an article on social media, take a step back and verify the publisher and check for obvious signs of veracity such as the type of users who have liked or commented. If you’re sure of misinformation spread, make use of the platform’s reporting tools.
- On Facebook, click on the three dots next to the post, and hit “Find Support or Report Post.”
- On Google, scroll to the bottom of the results, and hit “Send Feedback.”
- On Twitter, click on the three dots at the bottom of the feed and hit “Report Tweet.”
- On Instagram, click on the three dots next to the post, click “Report,” and then “It’s inappropriate.”
References and Recommended Reading
- Fake News Detection on Social Media: A Data Mining Perspective. Kai Shu, Amy Sliva, Suhang Wang, Jiliang Tang, Huan Liu.
- A Survey of Fake News: Fundamental Theories, Detection Methods, and Opportunities. Xinyi Zhou; Reza Zafarani.