Deep Learning to Improve Sentiment Analysis

Abstract

The amount of unstructured data on the internet is increasing with each passing day. People express their opinions on social media, blogs, and forums in different ways. Therefore, the need to process that unstructured data grows as well. The processing is important for extracting knowledge patterns from the data available online. Sentiment analysis is a popular technique that is used for knowledge extraction from people’s opinions. There are many businesses and product developers that use people’s opinions as a basis for improvement. We have seen great progress in Natural language processing and Sentiment Analysis techniques over the past few years. However, traditional sentiment analysis approaches focused on some particular type of data by using machines learning models and failed to achieve the best performance in terms of accuracy. There are many short comings in previous studies. Therefore, a major challenge is to overcome these problems to extract useful data from the huge amount of data that is available online. This thesis has been conducted to reduce the research gap by coming up with a better solution to improve the accuracy of the existing models used for sentence level sentiment analysis tasks. Thus, we have proposed a neural network-based sequence model (RNN-LSTM) that is used for the sentiment classification from opinionated sentences. Our Sentiment classification model is based on two state of the art deep learning algorithms Recurrent Neural Network (RNN) and (LSTM). We have evaluated our approach on four sentiment classification datasets. Furthermore, we have also made a detailed comparison with popular baseline approaches. The results prove that the proposed technique acheives improved accuracy as compared to the existing models.