Sentiment classification for movie reviews based on machine learning

The research of this article aims to achieve the classification of positive and negative emotions in film review texts. The dataset used is the IMDB dataset from the Kaggle website, which contains approximately 50,000 movie reviews that are classified as positive or negative emotions. This is a dataset for binary-sensitive classification containing substantively more data than previous benchmark datasets. This research will use two different models, SVM and CNN, to classify the dataset and then compare and analyze them. After completing the experiment, the SVM model achieved an accuracy of nearly 90%, while the CNN model only had an accuracy of about 82%. This research found that for binary classification problems, as long as the input data is processed well, SVM can also achieve results that are no less than or even better than CNN. SVM is simple and efficient but difficult to implement for large-scale training samples; CNN is suitable for solving problems with complex internal mechanisms, but its learning speed is relatively slow. Finally, we proposed our prospects for future research.