Impact of Feature Selection Techniques for Tweet Sentiment Classification

Sentiment analysis of tweets is a powerful application of mining social media sites that can be used for a variety of social sensing tasks. Common feature engineering techniques frequently result in a large numbers of features being generated to represent tweets. Many of these features may degrade classifier performance and increasing computational cost. Feature selection techniques can be used to select an optimal subset of features, reducing the computational cost of training a classifier, and potentially improving classification performance. Despite its benefits, feature selection has received little attention within the tweet sentiment domain. We study the impact of ten filter-based feature selection techniques on classification performance, using ten feature subset sizes and four different learners. Our experimental results demonstrate that feature selection can significantly improve classification performance in comparison to not using feature selection. Additionally, both choice of ranker and feature subset size significantly impact classifier performance. To the best of our knowledge, this is the first work which extensively studies feature selections effect on tweet sentiment classification.