Multi-Class Sentiment Classification using Feature Selection and Machine Learning Algorithms

Abstract

Sentiment analysis (SA) is a popular method for collecting relevant and arbitrary information from text-based data. To locate, examine, extract reactions, and emotions from the data or states, it applies computational linguistics, biometrics, text analysis, and Natural Language Processing (NLP). A SA model can be developed and improved with the use of the features analysis method. However, it can be difficult to find best classification methods for this type of data. When compared to current feature-based methods, machine learning approaches for analysis of sentiment, are capable of providing precise representation and improved performance. In this paper, the machine learning (ML) method was proposed for improving sentiment classification performance with a sophisticated embedding word method and develop an Emperor Penguin Optimization (EPO) and Multi-class Support Vector Machine (SVM). For the purpose of predicting the sentiment of tweets for classification, a large amount of Amazon data is analysed. This study, concentrate on improving sentiment analysis performance by building a Multi-class SVM network and particularized model of machine learning with novel word embedding techniques. The Multi-class SVM model has achieved 99.89% accuracy sentiment analysis.