Time effect in sentiment analysis

Abstract

Sentiment analysis is a trending topic that is widely studied in the recent years. As a result of social media being used actively, companies use machine learning systems to monitor and understand their customers' feedback. Over the time a given classical sentiment analysis system is being affected because of some phrases disappear and some other words emerging. On the other hand, a labeled dataset is required to analyse tweet data but because of finding labeled data is hard, active learning methods becomes the proper solution to the problem. Main aim of the active learning is to achieve same or better results with less training data. In this study, active learning methods are applied to two different tweet datasets using several different active learning methods such as clustering and choosing actively queried samples iteratively to investigate time effect. More accurate results are obtained by active learning methods according to the random selection.