Information Credibility on Twitter Using Machine Learning Techniques

Abstract

In today’s world, people are highly inclined towards social networking sites like Twitter, which provides a platform where users can have easy access to the high impact occasions/events rising worldwide. Users can share views and retweet the contents posted by other users with respect to such high impact event. However, users have diverse interests and hold strong opinions pertaining to any political party, caste, culture, religion, etc. So, while sharing any information, it is very essential for the social media users that they do not post any abusive or absurd content which might hurt the emotions of others and can end up into dreadful situations. So, there is a dire need of some filtering techniques to build a credibility analysis model which can filter out all such uncredible and questionable contents from social media. In this paper, a machine learning model has been developed to detect the credibility of tweets over four distinct credibility classes. Approximately 5k tweets were crawled from Twitter and preprocessed, which helped in developing a model. Initially, the K-Means clustering algorithm was applied on the dataset to find which tweet falls in which cluster, based on its similarity measures of feature set. The total variance in the dataset explained by K Mean Clustering Algorithm is found to be 86.6%. Furthermore, the Support Vector Machine algorithm is applied to build a classification model and classify the tweets into their respective credibility classes. It provides 96.53% accuracy with 99.51% area under the curve.