Feature Selection Methods in Sentiment Analysis: A Review

The development of digital tecnnologies nowadays assists people by suggesting opinion, choices, preferences and feelings. This opinion is useful for company's engagement to make certain analysis to know their potential users and personalized their need. However, the information needs extraction to make further analysis. Thus, sentiment analysis is used to extract opinion and others and transform it into meaningful data. During the process of analysis, feature selection method is required to select a subset which consists of relevant features to construct a predictive model. This process requires some conditions during the selection of feature subset. The required conditions for feature selection are that the selected feature subset must be small and relevant for a high dimensional dataset which considers the presence of noise plus there are no redundant features. However, some of the feature selection methods unable to fulfill all conditions. In this research, 40 papers were collected, classified and reviewed. We discussed on the feature selection methods in sentiment analysis based on its level of analysis and make comparison between these methods to know its limitation and advantages. The comparison made between methods are based on its accuracy and CPU performance. Finally, suggest the best/benchmark method for feature selection. The findings obtained from this research shows that hybrid methods obtain the best accuracy and CPU performance compared to the other methods.