Detecting Fake News with Tweets’ Properties

Abstract

Social media has replaced the traditional media and become one of the main platforms for spreading news [1]. News on social media tends to travel faster and easier than traditional news sources due to the internet accessibility and convenience. However, not all the news published on social media are genuine and/or came from unverified sources. False information can be created and spread easily through social media and this false news can potentially or deliberately mislead or misinform readers. The extensive spread of fake news brings negative impact to not only individual but also society [2]. Consequently, fake news may affect how readers perceive an online news on social media and indirectly mislead the way they respond to real news [2] [11]. Though there are some existing manual fact-checking websites developed to examine if a news is authentic, it does not scale with the volume of the fast spread online information, especially on social media. To overcome this problem, there are automated fact-checking applications were developed to tackle the need for automation and scalability. However, the existing application approaches lack an inclusive dataset with derived multi-dimension information for detecting fake news characteristics to achieve higher accuracy of machine learning classification model performance. To solve this limitation, we derived and transformed social media Twitter’s data to identify additional significant attributes that influence the accuracy of machine learning methods to classify if a news is real or fake using data mining approach. In this paper, we present the mechanisms of identifying the significant Tweets’ attributes and application architecture to systematically automate the classification of an online news.