A Systematic Survey on Deep Neural Networks for Sentiment Analysis

Sentiment analysis is a powerful approach for processing, analysing, speculating, and summarising unstructured content to convert it into reliable information. With the increasing demand for sentiment analysis in real-life applications, new discriminate machine learning and natural language processing techniques have been introduced in past years. However, the deep neural network is an emerging approach that learns by multiple layer feature dimensions of data and predicts faultless for large corpora with minimum computational cost. These neural networks provide the automatic feature engineering and advanced word embedding (TF-IDF, One Hot Encoding, Word2Vec, and Glove) for feature vector presentations that are helpful in extracting the meanings from complex linguistic texts. Therefore, a systematic survey is presented to provide detailed knowledge about deep neural networks that are rapidly applied in sentiment analysis with enhanced linguistic capabilities. The prime contribution of the existing researchers is highlighted with the focal point of deep learning-based sentiment analysis. Additionally, several benchmark datasets, feature engineering techniques, and existing deep learning-based research performance are also discussed. The prime objective of this survey is to present the credibility of the deep neural network to handle the sentiment analysis tasks.