DISCOVERING MOVIE REVIEW SENTIMENT CLUSTERS USING DEEP LEARNING
JU YEON SONG *
Provert Research Laboratory, 111 Charlotte Place Suite#100/Englewood Cliffs, NJ 07632, USA
*Author to whom correspondence should be addressed.
Abstract
Online movie reviews are influential consumer-generated content that provide criticism, analysis, and feedback for potential movie-goers to consider future movie purchases. Movie reviews also have been shown to have a significant impact on societal attitudes and public action. In our report, we attempt to quantify these complex trends using deep learning methods. Deep learning methods classify information by mimicking the structure of the brain or the organisation of neurons. In neural networks, sets of algorithms are designed to recognise patterns within a dataset. Each layer of the neural network is made of nodes, a place that is equivalent to a neuron. A node combines input of a data with a set of weights, assigning significance to inputs for the task the algorithm is learning.
The dataset of our report consists of the raw text data of 25,000 movie reviews from IMDB labelled as positive or negative. The deep neural network resulting from the dataset was trained via the word embedding method, a process specifically used for natural language processing problems such as machine translation. The accuracy of the training was 88.7%. Significant feature representations were discerned from our deep learning model, and the sentiments of movie reviews can be classified into 4 groups using the K-means method. It was found that these sentiments could be arranged into a spectrum between utmost praise and total abhorrence towards the movies.
Keywords: Deep learning, artificial intelligence, machine learning, neural network, training process, sentiment clusters