Detecting Misleading Information Propagation using Twitter Data

Abstract

Fake news detection on social media has assumed enhanced importance and is a trending research topics for data analysts around the world. Social media being used for news is highly risky since is cost effective and spreads at the speed of light. Also since social media is available at a very cheap cost via internet in India, everyone would rather access social media than go on a news portal and seek news. Ap parently it motivates a user to spread fake news and whether unintentional or deliberate a user ends up spreading fake news. While most of fake news may be aimed at commercial or personal gains, some of it may cause deliberate harm to a nation’s progress. Extremists may use fake news to spread misleading information propagation and carry out recruitment or funding for unwarranted activities. Twitter, as part of social me dia platforms (SMPs) is a leading portal being used by everyone for gaining access and giving out infor mation. In a recent research at Massachusetts Institute of Technology, it has emerged that fake news trav els six times faster than real news. Ironically, twitter provided data for research under the project. Leading the spread of misleading information propagation via fake news, twitter has emerged as one of the leading SMPs for fake news dissemination through user tweets and thus is only imperative that methods for fake tweet detection are improvised and perfected at the earliest to filter information. It is only after Twitter has established its wide reach that the problem at hand seems real and demands an ur gent solution. The sheer quantum of tweets makes it even more challenging to sort the news and remove anomalies. Various machine learning techniques have been devised to identify fake tweets. This report aims at studying prominent machine leaning techniques and subsequently innovate and suggest a customized methodology for detection of misleading information propagation on Twitter. Generally, researchers have identified detection of fake tweets as a classification problem but some unsupervised and unconventional learning models are also available. Major effort for any supervised learning technique is creation of data corpus. The paper highlights some techniques for the same. Consequently, inclusion of context as part of feature extraction has been achieved to align the intent behind the tweet with the text. As a first step towards the solution, a sentiment analysis of tweets from a set of users has been recommended details have been illustrated in subsequent sections. Various context based technique have been studied and implemented on data corpus and results produced have been compared. The report also studies various NLP techniques for data preprocessing and feature extraction before using classifiers to draw the outcome.