Rapid adoption of internet and social media platforms has led to rampant spread of information in near real-time. The explosive growth of social media (such as Facebook and Twitter) has allowed users to discuss, share ideas, and debate emerging issues including democracy, education, and health. News channels have evolved from newspapers, tabloids, and magazines to digital forms such as online news platforms, blogs, social media feeds, emails, and other digital mediums.
Facebook referrals account for 70% of traffic to news websites. However, this provides room to certain entities who distort the facts and news for furthering an agenda, creating biased opinions, manipulating mindsets, etc. This phenomenon is called fake news.
Fake news appears in forms like clickbait, propaganda, satire, parody, sloppy journalism, biased news, etc. Fake news has two parts. One of them is the content which includes either text, string content, images, and videos. Main characteristics include tone, grammar, and pragmatics. The other part is the context: when genuine content is shared with false contextual information.
One such recent example is of a supermarket store ALDI. During COVID-19, a fake post was circulated across Facebook, Twitter, and YouTube that showed footage of a large crowd rushing into an ALDI supermarket in Germany. It claimed that the video shows panic buyers storming the supermarket during the pandemic. However, the entire claim was false. The actual video was filmed in Germany in 2011.
Such kind of viral half-truth is part of the fabric of today’s internet. It triggers anger which turns into a dangerous commodity further exploited by businesses for their own benefits, by scammers for raising money, and by authoritarian governments for spreading hate. This calls for a reliable method to identify fake news and stop its circulation across different platforms.
How to fight fake news
Different methods are used to fight fake news. A few of them being:
1. Language approach: This includes use of linguistics by a human or software program to detect fake news. It considers different parts of the content: words in a sentence, letters in a word, structure, and the integration. More focus is on the grammar and syntax.
2. Topic agnostic approach: This approach focuses on detecting fake news by not considering the content of the article but rather topic agnostic features. It uses linguistics and web markup capabilities to identify fake news: for example, sifting through many advertisements.
3. Knowledge based approach: This approach is about integration of machine learning and knowledge engineering to detect fake news. It aims at utilizing external sources to verify if the news is fake or real and to identify the news before the spread becomes quicker.
4. AI based machine learning approach: Machine learning algorithms are useful in identifying fake news. It makes use of different types of training data sets to refine the algorithms. Datasets enable computer scientists to develop new machine learning approaches and techniques. These datasets are further used to train the algorithms to identify fake news. A simple framework built to combat fake tweets on Twitter focuses on these major areas: metadata of tweets, source of tweets, date and area of tweet, where and when the tweet was developed. By studying these four parts of the tweet, the given framework can be implemented to check the accuracy of data and separate the real from the fake.
Among several algorithms used here, some use the comparative analysis between similar posts to check if the information and facts are true and match up to the reliable sources. Others make use of the differences between the title and the content to identify clickbait articles. One example of such a methodology is what Facebook has deployed. It uses SimSearchNet++ which is an improved image matching model. It is trained using self-supervised learning to match variations of an image with a high degree of precision and improved recall.
Wipro has invested hugely in development of AI-based fake news detection solutions. We serve our clients with a two-way approach.
The first approach is tool-led. We have developed Vantage, an AI-powered in-house tool that performs analysis on the video and extracts all the key intelligence/meta data. It supports 125 languages. Vantage helps fact checkers, social media editors, and journalists monitor and spot fake news on social media platforms. The tool can easily verify the piece of information, identify the main source of the content, check the reliability of the source, and determine if the content can be shared and published.
The second approach is for scenarios like deep fake where system-led detection is not accurate. Such cases are passed on to human agents for manual verification. The human intervention is finally utilized to train the system. The combined efforts have benefitted our clients in internet and social media domains, improving their online experience and reducing the influence that fake news may have on the end users.