Sentiment Analysis of Movie Reviews with Deep Learning Methods

Abstract

none

Sentiment analysis is one of the most popular and important research field in natural language processing (NLP). The purpose of this thesis is to propose a deep learning neural network for polarity sentiment analysis of movie reviews. Preparation data is the foundation to build the sentiment analysis model. In this phase NLP techniques will be useful. Preprocessing for the data has been implemented in this work. In this study, we focus to measure semantic similarity between words and the system will learn word embedding by the data for fitting the neural network to create a sentiment analysis classification model of movie reviews which can predict the outputs of positive or negative opinions on the documents. Our experiment is to creates 5 models of neural networks for comparison to achieve a better result. Long-Short Term Memory (LSTM) is used because the memory cell can memorize the long term of words, and carry the previous information to current input. Furthermore, Bidirectional LSTM (BLSTM) is used which can carry information from the past and the future. Besides, Convolutional Neural Network (CNN) is also experimented in this study. We make a comparison between the networks of single LSTM, BLSTM, CNN-LSTM, CNN-BLSTM and CNN. Finally, we have successfully to achieve a high accuracy for this study. BLSTM network achieves the best performance of accuracy (89.39%) and F1 score (89.99%).