Sentiment Analysis Using Ensemble of Deep Learning Models

Sentiment analysis (SA) using machine learning (ML) is the common practice to detect the sentiments. During critical happenings such as social activities or natural calamities, SA model is not capable to detect association between words, vital qualitative historical information, and social impact related to events which are considered as main features. In SA or opinion mining (OM), we comprehend people’s reactions, views, opinions, and outlooks stated in written languages. SA is actively researched in the domain of text processing and natural language processing (NLP) in the recent past. SA can be applied widely because views or opinions are frequently expressed by the humans, offers knowledge for decision-making. In addition, it presents several challenges in detecting efficient sentiments, thus, task based on SA is spread in management, business, social sciences, and computing sciences due to its impact in society as well as business. In this article, we propose improvised deep model by using Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) to detect the sentiments from Twitter dataset. The ensemble of CNN and LSTM is proposed for better feature extraction and improved results. Training accuracy of 99% and validation accuracy of 95% is achieved by the developed model. Moreover, the comparative analysis shows that the proposed model is superior to state of the arts. Thus, we suggest the use of CNN and LSTM for SA to accomplish better performance metrics.