Credibility Detection on Twitter


University of Cape Town Computer Science Honours Project


Introduction and Background

The role of social media in our day to day lives has increased rapidly in recent years. It is now used not only for social interaction, but also as an important platform for exchanging information and news. Twitter, a micro-blogging service, connects millions of users around the world and allows for the real-time propagation of information and news.

Twitter has, however, not only been used for the spread of valid news, but also deceptive and fake news. This fake news can come in the form of spam, astroturf (a technique used in political campaigns to fake support numbers, by making a message appear to have `grassroots' origins when in reality it originated from one person or organisation), clickbait (content that aims to attract attention and get users to click on a link to increase website traffic) and more. The increase in the volume of fake news has even led to our current times being labelled `the age of misinformation' and therefore stresses the importance of assessing the credibility of tweets.

Problem Statement

Our project focuses on developing a system that can algorithmically assess the credibility of tweets on Twitter, and present the assessment results to the user. A classifier will be trained using an annotated data set, generated through a crowdsourcing mobile application, and results will be displayed in the Twitter interface via a Web browser extension.

Project Goals and Significance

The existence of fake news is not new. Before the use of social media, news was restricted to sources such as the radio, newspapers and TV, where the task of filtering out fake news was assigned to journalists and other news publishers. The rapid increase in user generated content has, however, meant reliance on these traditional filtering techniques is no longer applicable. Research has found that humans are not good at detecting lies in text based on content alone and so there has been a drive to automate news credibility evaluation.

We hope our project helps to improve people's critical judgment of online news and encourages people to be more aware of its existence. We hope to reduce the spread of fake news and aid people in their credibility assessment of the content that they consume.

Project Breakdown

The project was broken up into 3 main sections. These sections come together to form the credibility checking system, as seen in the diagram below. The three sections are:

  1. A mobile crowd sourcing application. This is used to collect a dataset of annotated tweets that can be used to train the tweet classifier
  2. A multiclass machine-learning classifier. This uses a set of tweet related features to calculate a tweet's credibility rating.
  3. A Chrome Web browser extension. This displays the tweet credibility ratings to the user by augmenting the Twitter interface.
System Diagram

Meet the Team

Shaheen Karodia

Shaheen Karodia

Crowdsourcing Application Developer

Michelle Lu

Michelle Lu

Trained a Machine-Learning classifier

Kristin Kinmont

Kristin Kinmont

Web Extension Developer