Sentiment analysis has recently drawn considerable research attentions in the recent years owing to its applicability in determining users opinion, sentiment and emotions from large collections of textual data. The goal of sentiment analysis centered on improving users experience by deploying robust techniques that mine opinions and emotions from large corpora. Although there are a number of studies on sentiment analysis and opinion mining from textual information, however, the existence of domain-specific words such as slang, abbreviations and grammatical mistakes further posed serious challenges to existing sentiment analysis methods. Therefore, research efforts have focused on finding the most discriminative attributes that can help in capturing users opinions from textual datasets. In this paper, we focused on identification of effective discriminative subset of features that can aid classification of users opinion from large corpora. This study proposed hybrid feature selection framework that is based on hybridization of filter- and wrapper-based feature selection methods. Correlation feature selection (CFS), a filter-based approach is hybridized with Boruta and Recursive Feature Elimination (RFE), which are wrapper-based feature selection methods, to identify the most discriminative features subsets for sentiment analysis. Four publicly available datasets for sentiment analysis: Amazon, Yelp, IMDB and Kaggle were considered to evaluate the performance of the proposed hybrid feature selection framework. This study evaluated the performance of three classification algorithms: Support Vector Machine (SVM), Naïve Bayes (NB) and Random Forest (RF) to ascertain the superiority of the proposed approach. Experimental results across different contexts as depicted by the datasets considered in this study clearly showed that CFS combined with Boruta produced promising results especially when the features selected are passed to RF classifier. Indeed, the proposed hybrid framework provide effective way of predicting users opinions and emotions while giving substantial consideration to predictive accuracy
Key words: Sentiment analysis, opinion mining, hybrid feature selection, Boruta, Recursive feature elimination
|
