Sentiment Analysis of Movie Reviews based on Machine Learning
Yang Zhao · 2020
Nowadays, people are overwhelmed by fragmented text information online. It is easy for a person to post comments on a movie without actually watching it. Therefore, analyzing the sentiment of the movie reviews becomes necessary and significant. We first got the data of movie reviews and transformed the text data into vector data by using the TF-IDF algorithm. The total amount of the transformed features was approximately 72,000 to describe the sentiment of movie reviews from different people. Three machine learning models were used to train by these feature vectors, which were Lasso Regression (L1 Logistic Regression), Ridge Regression (L2 Logistic Regression), and CatBoost models. Finally, the evaluation of these three methods showed that the accuracy and the precision of Lasso Regression were only close to 0.8. It was the worst prediction model among these three models while the accuracy and precision of CatBoost were more than 0.85, which was the best. This paper not only helps guests understand the accurate evaluation of movies, but also helps filmmakers predict the box office performance by comparing the performance of these three machine learning models in terms of sentiment analysis.