Sentiment Polarity Detection from Amazon Reviews: An Experimental Study

Abstract

With the ever increasing number of electronic commerce portals and the selling of goods on the Web, customers’ reviews are usually used as means to grasp the goodness of products. Mining and understanding the polarity of reviews is therefore crucially important for future customers that seek opinions and sentiments to support their decision buying process. This paper proposes an experimental study of SentiME, our approach for extracting the sentiment polarity of a message, on the Amazon review-based corpus provided by the ESWC SSA challenge. We use an Ensemble Learning algorithm implementing five state-of-the-art classifiers that are known to well perform in the domains of tweets and movie reviews. The Ensemble Learning is trained with a Bootstrapping Aggregating process using a set of linguistic (such as ngrams), and semantics (such as dictionary-based of polarity values for emojis) features. The approach presented in this paper has been first successfully tested on the SemEval Twitter-based corpora. It has then been tested in the ESWC Semantic Sentiment Analysis 2016 challenge, where properly trained, it reaches a F-measure of 88.05 % over the test set for the detection of positive and negative polarity, which ranks our approach as the first system among the ones competing in this challenge.