Opinion Mining of Politics and Inflation using Roman Urdu Dataset
Zunaira Shafqat, Muddesar Iqbal, Waqas Haider Bangyal, Dhafer Almakhles
Abstract
In today's world, data volume is immense and is expanding at an exponential rate. A leading factor of this is the data shared over social media by millions of people daily in the form of text, audio, video, and images. Opinion mining is a branch of text mining that seeks to extract public opinion about an event or particular topic on any online platform like review sites, Twitter, Facebook, etc. Most of the work in opinion mining is done in the English language and very little is done in Roman Urdu, which is a dominant social media language in Pakistan and India. Inflation and politics are amongst the most talked-about and discussed topics on social media in Pakistan. In this study, an Inflation and Politics-based Roman Urdu dataset is prepared that is extracted from the Roman Urdu dataset available at Kaggle. Various operations (with text processing, without text preprocessing, with attribute selection, without attribute selection) were performed on the data over machine learning algorithms of Naive Bayes, Bayes Net, KStar, Decision Tree, and Random Forest and determined which algorithm gave the best accuracy on the training data. WEKA (Version 3.8.4) was used to generate opinion mining regarding politics and inflation in Pakistan using the Roman Urdu dataset.