Bayes Rule and Fake News
Fake news is the deliberate misinformation via print, broadcast, or social media. This is an increasing problem going on in today’s political climate. This problem is so prevalent that the president of the united states address it whenever he is being slandered. Due to the increase in social media use today the form of fake news is beginning to change. It is becoming increasingly difficult to distinguish what is fake news and what is real news. Fortunately, social media platforms have a method of counteracting fake news.
https://www.kdnuggets.com/2017/04/machine-learning-fake-news-accuracy.html
According to the author of the article, Facebook (and assuming other social media platform) use AI to find fake news and remove them from their platform. This AI uses Bayes rule (specifically Bayes classifier) to classify and differentiate fake news and real news. The author of this article attempts to replicate the AI. While the actual model is much more complicated, the author basically created a dataset of fake news articles counted the frequency of each word in each article and divided it by the amount of words in the article. He, then, uses the most common words and phrases found in the dataset, thus finding P(wi|F). Using the most common words in fake articles, he found the probability of those words in the dataset of real news to finding P(wi|R). He founded the probability of an article being real (p(R)) and the probability of the article being fake (P(F)). Then, to determine that the article is fake, the author goes through each word and find the probability of the article is fake given that frequency of the words in the article (the author used an overall Bayes rule (P(F|w) = multiplication of((P(wi|F)*(P(F)))/multipication of(P(wi|F)*(P(F)) + multiplication of(P(wi|R)*(P(R)). If the P(F|w) is greater than .5, the article is most likely fake.