A pragmatic analysis of text-based sentiment analysis models from an analytical perspective

Abstract

NLP sentiment analysis involves pre-processing, feature extraction, feature selection, sentiment classification, and post-processing. Pre-processing can include PoS tagging, stopword removal, language-specific missing work detection and correction, etc. Word2vec, N Gramme, term frequency (TF), distance measures, etc. can extract and select features. CNNs, DNNs, and other machine learning and deep learning models can be used for classification and post-processing. These models form a sentiment analysis engine, which can be assessed for accuracy, precision, recall, analysis delay, computational complexity, etc. The wide variation in algorithmic combination and performance makes it difficult for researchers to choose the best model(s) for their application. This task requires a lot of model testing and validation, which increases time-to-market and product cost. This text surveys recently proposed sentiment analysis models to reduce ambiguity and speed up system development. This survey characterises reviewed models’ nuances, advantages, limitations, and future research scopes to help readers choose the best models for their application deployment. These models’ delay, accuracy, precision, recall, computational complexity, and probable use are also compared in this text. This comparison lets readers choose the best model(s) for their application deployment and use them without validation. Reducing testing delay will improve model performance for system design and deployment for multiple sentiment analysis applications