Analyzing Twitter Data for Text Classification

Abstract

Predicting the polarity of words and subsequently classifying them into positive or negative sentiment is the major
emphasis of sentiment analysis, a classification issue. The two most common kinds of classifiers are those that rely
on a vocabulary and those that employ machine learning. Word Sense Disambiguation and SentiWordNet are
examples of the former, whereas RNN Classifier, Multinomial Naive Bayes (MNB), Logistic Regression (LR),
Support Vector Machine (SVM), and others are examples of the latter. In this paper, we make use of two preexisting
datasets: one from Stanford University's "Sentiment140" with 1.6 million tweets, and another from CrowdFlower's
Data for Everyone library with 1,837 entries; both datasets have already been classified according to the sentiments
conveyed within them. We compare the results achieved by the following sentiment classifiers—Textblob,
Sentiwordnet, MNB, LR, SVM, and RNN—using the aforementioned dataset to categorize tweets as positive or
negative. In addition to the aforementioned machine learning methods, the datasets have also been subjected to an
ensemble version of MNB, LR, and SVM. In addition, you may utilize the learned models mentioned earlier to
forecast the sentiment of fresh data.