Date of Publication :30th September 2020

Abstract: Sentiment analysis [6] has long been an important problem of study in NLP and machine learning, for finding public sentiments on products, brands or services. Previous approaches to sentiment analysis have included unsupervised learning, Naive Bayes classifiers and SVM. In this paper, we focus on sentiment analysis of movies using text reviews. Sentiment analysis can be a challenging problem to solve because our language is rather complex and a single word can have either positive or negative connotations based on the context. We will be using the Large Movie Review Dataset [4] given by Stanford AI lab, which is a binary sentiment classification dataset based on IMDB reviews of movies, and contains 50,000 reviews with a 50:50 train: test split. The objective is to classify a movie as good or bad, based on its text review. We will approach this problem using the Tf-Idf vector of the corpus and applying a deep learning model on top of it. This model achieved an accuracy of 90.7%, which is a significant improvement over the current approaches. Future extensions to this approach could include more powerful deep learning models like LSTM or GRU, which can extract even more contextual information.

Reference :