Twitter Data Classification by Applying and Comparing Multiple Machine Learning Techniques

Abstract

Having an average of five hundred million  tweets sent out per day, twitter has become one of the  largest platforms of data analysis for the researchers.  Previously, various researches have been conducted on  twitter data i.e., sentimental analysis. However, not  much research has been done to classify the tweets in  terms of categories so that tweets can be distributed as  per user preferences. In this research we started by  creating four broad categories: politics, sports, crime  and natural. After that, we applied different machine  learning techniques (Random Forest, K-Nearest  Neighbors, Naïve Bayes, Logistic Regression, Decision  Tree and Support Vector Machine) to classify the  twitter data. Finally, we compared the results in terms  of sensitivity, specificity, precision, false positive rate  and accuracy. We found that Support Vector Machine  (SVM) produced the best results in terms of sensitivity,  specificity, precision, false positive rate and accuracy.  Hence, we concluded that a machine learning approach  (Support Vector Machine) can certainly be used to  classify twitter data. Constructed dataset, all the  programs, figures and snippets can be found at  https://github.com/ananyasarkertonu/Twitter-Dataset