Tweeple's microblogs on illegal immigration in USA
Sultan Menwer Altarrazi, Sreela Sasi · 2016
Opinion mining has become the center of attention for many researchers and scientists. That's because people share their thoughts and opinions as texts precisely in microblogs such as on Facebook and Twitter. This research presents a process for opinion mining of Tweeps, the people who tweet on Twitter. The immigration topic was chosen specifically in comparison with other important topics of politics because it has remained as an unsolved problem for decades. The data for this research was collected after the US Republican Presidential election debate on October 28, 2015 at University of Colorado in Boulder for a period of approximately 10 days. The three major categories of opinion identified are `reform or give citizenship to illegal immigrants', `deport all illegal immigrants', or `deport only the criminal illegal immigrants'. In a manual walkthrough of the data, the majority of the opinions were found to be biased towards the second category, which is `deporting all illegal immigrants'. Binary classification for first two opinions (for and against) and multinomial classification for all three opinions are done. Random Forest, Multinomial Naïve Bayes, Linear Support Vector Machine, and Logistic Regression classifiers were used for this purpose. The results obtained using all four classifiers for both the pathways were strongly and dramatically convincing. Linear Support Vector Machine and Ensemble approach using Random Forest classifiers showed better accuracy for each class, and lower percentage error.