Document-level analysis of sentiments for various emotions using hybrid variant of recursive neural network

Abstract

Sentiment analysis makes use of natural language processing, computational linguistics, and analysis of text in order to extract information from text. In this project, we consider machine learning algorithm for classification of longer sentences with polarity from a huge amount of data from articles, forums, consumer reviews, surveys, blogs, Twitter, Whatsapp chat, etc., and come out with suitable results sentiments (e.g., sad, happy, surprised, angry) and give intensity/degrees. Naïve Bayes and SVM are complex and inaccurate and cannot handle large amount of data. Recurrent neural network takes into consideration the sequence of words in a sentence. We use a hybrid of both LSTM and GRU to learn the long-term dependency and do sentiment analysis. Our end goal is to perform sentiment analysis on long sentences and get accurate visualization results. Sentiment analysis can be widely applied to social media and reviews for various applications, such as customer service, marketing.