Feature Selection And Optimized Weight Based Multi-Tier Stacked Ensemble (Owmtse) Classification For Twitter Sentiment Analysis

Abstract

In recent years, a huge number of people have been attracted to social-networking platforms like Facebook, Twitter and Instagram. Most use social sites to express their emotions, beliefs or opinions about things, places or personalities. Sentiment Analysis (SA) is important to improve particular products (or) topics. There has been lot of work in the field of sentiment analysis of twitter data. The previous system designed an Optimized Weight based Multi-Tier Stacked Ensemble (OWMTSE) framework for twitter sentiment analysis. However, the extracted features may cause a complex computation problem due to overfitting. To deal with such problem, optimal feature selection techniques are required.  To solve this problem the proposed system designed a Binary Swallow Swarm Optimization (BSSO) based feature selection and Optimized Weight based Multi-Tier Stacked Ensemble (OWMTSE) classification for Twitter Sentiment analysis. Initially, each tweet is represented by a vector of numbers based on the extracted features. Then the optimal features are selected by using Binary Swallow Swarm Optimization (BSSO) algorithm. Finally, the OWMTSE learning is designed for sentiment classification.  In the proposed OWMTSE system, Weighted Majority Voting (WMV) ensemble classifier with Support Vector Machine (SVM), Convolutional Neural Network (CNN) and Passive Aggressive Classifier (PAC) in as base learners. This novel algorithm uses the incremental learners to predict the results that get combined by the classifiers in the next tier. The meta-learning in the next tier generalizes the output from the classifiers to give the final prediction.  Results of these classifiers are tuned via the optimized weight via Genetic Algorithm (GA). The voting ensemble model will consider this tweet as positive one because this is the majority decision. The experimental results shows that the proposed system attains higher performance compared with the previous methods interms of accuracy, precision, recall, f-measure and error rate.