Negation Handling using Stacking Ensemble Method

Abstract

Polarity shift is the major problem in the Bag-of-words model. Polarity shifting occurs when the polarity of the sentence is different from the polarity expressed by the sum of the content words in the sentence. Polarity shift reverses the sentiment polarity of the text. It affects the classification performance of the machine learning algorithms. Negation, contrast and sentiment inconsistency are the three different types of polarity shifts. The proposed system consists of four modules namely preprocessing, polarity shift detection and elimination, sentiment classification, and stacking ensemble method. Polarity shift detection is a hybrid approach. It is a combination of rule and statistic based method. Linear Support Vector Machine, Logistic Regression and Naive Bayes are used for sentiment classification. The Weka tool is used for the implementation of the machine learning algorithms. Stacking ensemble method combines the output of the base classifiers and gives the integrated model. The paper focuses on the document level sentiment analysis. Stacking ensemble method helps to increase the accuracy of the machine learning algorithms. The system is analyzed using airline reviews. Airline reviews are taken from Skytrax website. Reviews are categorized into four types namely airline, airport, lounge, and seat. Best airline of the year can be identified by the proposed system.