Exerting 2D-space of sentiment lexicons with machine learning techniques: A hybrid approach for sentiment analysis

Abstract

Sentiment mining from the textual content on the web can give valuable insights for discernment, strategic decision making, targeted advertisement, and much more. Supervised machine learning (ML) approaches do not capture the sentiment inherent in the individual terms. Whereas the unsupervised sentiment lexicon (SL) based approaches lag behind ML approaches because of a bias they have towards one sentiment than the other. In this paper, we propose a hybrid approach that uses unsupervised sentiment lexicons to transform the term space into a twodimensional sentiment space on which a discriminative classifier is trained in a supervised fashion. This hybrid approach yields higher accuracy, faster training, and lower memory footprint than the ML approaches. It is more suitable for scenarios where training data is scarce. We support our claim by reporting results on six social media datasets using five sentiment lexicons and four ML algorithms.