A Search for Optimal Feature in Political Sentiment Analysis

This research used the data from Twitter on presidential elections in USA 2016 to understand which features better suits in predicting Election results. We compare between four features such uni-grams, bi-grams, tri-grams, and opinion words by using the data mining techniques such as Random Forest, Naïve Bayes, and Artificial Neural Network. For the analysis, we have used a Data set from Kaggle consisting of 6445 individual records. Then applied many preprocessing techniques (such as cleaning, data stemming, data normalization etc.) on the said data set to expose the well-shaped data set. For extracting proper features, we have done factor analysis. Finally, we have tested our method using the dataset and found the uni-gram showing the better accuracy of 81%. This research was implemented in R.