A Comparative Study of Evolutionary Methods for Feature Selection in Sentiment Analysis

Shikhar Garg, Sukriti Verma

Abstract

With the recent surge of social media and other forums, availability of a large volume of data has rendered sentiment analysis an important area of research. Though current state-of-the-art systems have been demonstrated impressive performance, there is still no consensus on the optimum feature selection algorithm for the task of sentiment analysis. Feature selection is an indispensable part of the pipeline in natural language models as the data in this domain has extremely high dimensionality. In this work, we investigate the performance of two meta-heuristic feature selection algorithms namely Binary Bat and Binary Grey Wolf. We compare the results obtained to employing Genetic Algorithm for the same task. We report the results of our experiments on publicly available datasets drawn from two different domains, viz. tweets and movie reviews. We have used SVM, k-NN and Random Forest as the classification algorithms.

Download


Paper Citation