A Novel Hybrid Lexicon Ensemble Learning Model for Sentiment Classification of Consumer Reviews

In recent past during the era of consumerism with easy accessibility to social networking world, the consumers usually give comments and opinions on daily usable ingredients, electronic goods and services offered on payments. These comments or opinions are innumerable and huge on each item, hence need the special attention for sentimental value particularly on their text parts. The present study is an attempt to perform sentiment prediction on Amazon Electronic products, gift cards and Kindle dataset. In this paper, the HLESV (Hybrid Lexicon Ensemble based Soft Voting) model is proposed by combining lexicon and ensemble approaches using optimally weighted voting to predict the sentiment, subsequently to evaluate model using various performance metrics like precision, recall, F1-score. This paper computes an additional metric namely subjectivity score along with sentiment score and proposes non-interpretive sentiment class label to evaluate the polarity of the reviews using our proposed HLESV model for sentiment classification. The accuracy score of our proposed HLESV model is evaluated to assess its effectiveness on Amazon consumer product review datasets and observed an increase of 1-6% accuracy over existing state-of-the-art ensemble methodology.