Sentiment Analysis
Abstract
Sentiment analysis is the problem of determining the attitude or feelings atext tries to convey. It has recently received a lot of focus, one reason beingthat the amount of text containing personal opinions has increased in stepwith the emergence of social media.We implement and evaluate a supervised, unsupervised and a hybrid methodfor classifying the sentiment of a document. We find that for the super-vised method, the most important aspect is how we select which featuresto use, and which to discard.Out of the systems we evaluated, the supervised method clearly performedthe best. The Na ̈ve Bayes algorithm managed to get the best performanceıon our test sets, with an accuracy of 98.3% on one test set, and 90.5% on amore difficult set.