Decision tree combined with PSO-based feature selection for sentiment analysis

Abstract

Sentiment analysis can be considered as a classification task in natural language processing as it harnesses classification algorithm to predict a particular class in a text data. In the classification task, feature extraction is a process to extract the features of the data so that it can be used as the input of the classification algorithm. However, not all features are particularly relevant for a classifier. Irrelevant features might significantly decrease the performance of classification algorithm. This paper proposes a PSO-based feature selection, combined with decision tree algorithm (PSO-C4.5) for sentiment analysis. The PSO-C4.5 is validated on a private data set, which is a sentiment data set about online transportation in Indonesia. The proposed method considerably enhances the performance of decision tree in comparison with the baseline. ?? Published under licence by IOP Publishing Ltd.