A Pragmatic Approach to Emoji based Multimodal Sentiment Analysis using Deep Neural Networks

Abstract

The Opinions of the customers regarding products have also become an important parameter for sales. The manufacturing companies are also continuously monitoring the feedback given on social media sites about their products, especially mobile reviews. The Sentiment Analysis (SA) is playing a vital role. The analysis cannot be limited to only text categorization as positive, negative, or neutral. The Emojis are also capturing emotions. So, in our proposed work the multi-modal sentiment analysis is done using text and Emojis. And the malleability of Deep learning models on the text has also increased. The combination of Word embedding models CBOW and SG are combined with the deep learning classifiers like LSTM, CNN, Bi-LSTM, and CNN-LSTM. The novelty of this work is to develop an Emoji-Based sentiment lexicon and cosine similarity usage for finding similarity. These were all modeled to predict the emotions in new mobile product reviews collected from various social media sites. The evaluation parameters proved that our proposed work had better results. The CNN-LSTM model topped in the accuracy of 94.94%.