Since the inception of social media, the problem of fake news has spread like wildfire. Fake news can affect politics, healthcare, the tech industry, and much more. If someone tells you that there is a disease spreading through onions, it can cause a decrease in sales of onions, affect people's lives and lead to a huge wastage of food. That's how big fake impact news can have on our lives.
So, how can we prevent falling for fake news? By not reading news, just kidding! This is where AI comes into the picture. Tech giants like google are using artificial Intelligence, Facebook to detect and avoid the spread of such articles. Let's discuss in brief the methods used for the classification of news articles.
1. Score based method
This method involves the following steps
Use the accuracy of facts presented to score the web pages
Elicit the meaning from different parts of the article and judge whether they are related or make some sense.
The reputation of the source will adobe a major factor contributing to the judgment. A reputable source can be trusted.
These words are often used to draw the attention of the readers but can be used by AI algorithms to predict the genuineness of the article.
2. Using Generative Adversarial Networks
Generative Adversarial Networks can generate fake news articles within seconds, and these articles generated are similar to the real ones. GANs use two networks, a generator and a discriminator. While the generator generates articles, the discriminator tries to identify them for authenticity, constantly improving throughout the process. As the training progresses, both the networks become better at their jobs, and now a GAN can generate a fake article very similar to authentic ones, but how does this help us? It is helpful because the discriminator from this network can identify fake articles, especially those generated using AI.
3. Using GLTR
Harvard developed the Giant Language model Test Room; the main idea behind GLTR is to use statistical analysis to detect text generated using AI. Let's understand how it works, and now we know that machines generate text based on the training set, which is enormous and follows a distribution in the usage of words. We now use the text and compare it with standard distribution, and if multiple words follow a distribution pattern, the chances are high that the text is machine-generated.
These were some methods that are commonly being used to detect fake news articles. Although these methods are not perfect, the progress with accuracy over time is high; we can say that soon we will be able to completely differentiate fake and real articles.