Meta Classifier-Based Ensemble Learning For Sentiment Classification

Naznin Sultana, Mohammad Mohaiminul Islam

Abstract: In the era of digital explosion, a huge amount of data is generated every second from different sources, which requires rigorous analysis for decision-making and knowledge gathering. Recently, one such challenging issue that found to be interesting to many researchers is sentiment classification through opinion mining. From the latest research on sentiment analysis, it has been also observed that ensemble-based supervised machine learning algorithms outperform individual classifiers. Ensemble methods are meta algorithms that combine several machine learning algorithms into one predictive model, and thereby decreases variance and bias and improve classifiers prediction capability. In this paper, we have evaluated the performance of ensemble methods in conjunction with meta algorithm on two different datasets. Majority voting algorithm has also been employed for prediction decision of sentiment classification for our model. The experimental result shows that meta classifier-based ensemble approach outperform individual classifier in some cases.

External IDs:dblp:conf/iwci/SultanaI18