Building Large Scale Cloud System for Product Sentiment Analysis using Hybrid Group Search Optimization Based Feature Selection

Abstract

A very powerful technology that performs complex computing in a massive scale is known as Cloud computing. There has been a massive growth that has been observed in the data scale which may also be big data which is generated by means of cloud computing which is observed. Sentiment Analysis, on the other hand, denotes the opinion extraction of users from the documents used for review. A sentiment classification that makes use of methods of Machine Learning (ML) can face problems in high dimensionality for a feature vector. Thus, the method of feature selection is needed for the elimination of all noisy and irrelevant features from a feature vector for efficiently working the ML algorithms. All chosen features will be sub-optimal owing to a Non-Deterministic Polynomial (NP) hard type of technique that was used. The Group Search Optimization (GSO) based algorithm which was on the basis of a method of feature selection will find some optimal feature subsets through the elimination of all redundant features. For this work, the method of feature selection based on the GSO was applied to the sentiment classification. There was also a method of feature selection which was hybrid and based on the GSO and Local Beam Search (LBS) that has been proposed for a sentiment classification. The methods proposed were evaluated based on the product review dataset of Amazon. The results of the experiment proved that this method of a hybrid feature selection can outperform all other methods of feature selection for a sentiment classification.