Twitter sentiment analysis offers organizations an ability to monitor public feeling towards the products and events related to them in real time. Most existing researches to identify Twitter sentiment are focused on the extraction of new sentiment features and apply pre-processing before features selection, although ignore the role of tweet pre-processing. In this paper, we discuss the effects of pre-processing on sentiment classification performance. We evaluated the effects of URL, stopword, repeated letters, negation, acronym and number on sentiment classification performance using two feature models and four classifiers on five Twitter datasets. The experiments show that the accuracy of sentiment classification rises after expanding acronym and replacing negation, although hardly change when removal URL, removal numbers and removal stopword are applied. The various pre-processing methods cause different influence on performance of classifiers for each dataset.
展开▼
