Sentiment analysis is one of the most popular research topics in last years. There are lots of data on web which require analysis in order for them to become useful. Many researchers have focused on making sense of these data. Therefore, sentiment analysis concept is proposed. Sentiment analysis methods try to emerge any opinions, feelings, and subjectivity behind the text. Machine learning algorithms and vocabulary based methods are used to perform sentiment analysis. In this research, (i) recently studied researches on machine learning based sentiment analysis are investigated to give background; (ii) they are classified according to their tasks on extracting information; (iii) the encountered and potential challenges on this research topic are revisited and discussed.