AN OPTIMIZATION OF POLARITY CLASSIFICATION FOR SENTIMENT ANALYSIS USING METAHEURISTIC FITNESS FUNCTION

Abstract

Opinion Mining (OM) has a big impact in text mining applications regarding customer attitude identification, brand/product positioning, customer relationship management as well as market research. It is also referred as sentiment analysis and is a natural language processing method for tracking public moods regarding particular products or topics. OM constructs a model for collecting and examining product opinions in comments, tweets, blog posts or reviews. The aim is to rank opinions as very bad , bad , and average, good , very good and so on opinions are rated between 1 to 5. newlineIn the initial level of research, a weighted semantic feature expansion utilizing hyponym tree for features integration is proposed. To improve the efficacy of the classifiers, a decision tree based features selection is suggested. The proposed techniques were validated using the popular Internet Movie Database (IMDb) dataset for sentiment analysis and medical dataset, created from popular blogs. Multilayer Perceptron (MLP) Neural Network (NN) is used for classifying the selected features as positive or negative. As small networks have limited information processing power it might not provide good performance. In the proposed algorithm, the Back Propagation (BP) parameters namely learning rate and momentum are optimized using Genetic Algorithm (GA) and proposed hybrid GA (HGA) along with the structure of NN. The MLP-HGA method increased classification accuracy significantly.