Sentiment Analysis Using AI: A Comparative Study Comparative Study of 5 Different Algorithms and Benchmarking Them with A Qualitative Analysis of Training time, Prediction time, and Accuracy

Abstract

Sentiment analysis also called Opinion Mining has been one of the most ancient topics in Natural Language Processing (NLP). NLP which is a subfield of AI enables machines to read, understand, interpret and manipulate human languages. The netizens often express their sentiments through tweets, reviews and ratings. Therefore, it's becoming increasingly popular to analyse these sentiments for commercial applications such as market analysis, product reviews, customer services and many more. This analysis can either be lexicon-based or machine learning-based. Lexicon-based techniques use dictionaries of words, which is then used to calculate a score for the polarity. Machine learning-based classifiers use a dataset to train a model which is used to find the sentiment of a sentence or a paragraph. ML-based approaches are more robust compared to the lexicon-based approach. The main contribution of this paper is to demonstrate the performance of various Machine Learning and Deep Learning models in Sentiment Analysis. In this paper, various models viz. Multinomial Naive Bayes(MNB), Support Vector Machine(SVM), Hidden Markov Model(HMM), Long Short Term Memory(LSTM) and Bidirectional Encoder Representations from Transformers(BERT), are trained through supervised learning, and then they are benchmarked through a qualitative analysis of accuracy, effectiveness and speed to provide an extensive overview of the various algorithms used in the field. The performance comparison of the various models presents an in-depth understanding of the different techniques within the framework of a comprehensive evaluation and discussion.