Intelligent Analysis System of Movie Reviews Using Deep Learning and Convolutional Neural Networks
Dingyi Yu · 2021 IEEE Conference on Telecommunications, Optics and Computer Science (TOCS) · 2021
Along with the rapid popularization of electronic products, various social media such as Twitter and Microblog provide people with more platforms to express their opinions. To effectively extract and classify the sentiment under such a massive number of opinions, usually expressed by plain text, has aroused people’s closed attention from the perspective of data analysis. With the development of deep learning algorithm, natural language processing model has been enhanced to achieved impressive success on opinion mining and text classification. In this paper, we retrieved 25,000 movie reviews with their corresponding sentiment labels and attempted to build a sentiment analysis system of movie reviews. We improved the traditional Bag of Words (BoW) model with TF-IDF algorithm to vectorize the data and constructed Convolutional Neural Network (CNN) training algorithm to detect people’s outlook on the movie whether they have positive or negative feeling. The final model is evaluated with average accuracy, average stability, and separate performance on detecting positive and negative sentiment. The best model is shown to have 80.62% average accuracy, with 1.33 standard deviation and 76.45%, 84.66% accuracy on detecting positive and negative sentiment, respectively.