Improving and Analyzing the Movie Sentiments Using the SVM Approach

Abstract

Sentiment analysis is a tool that assists in better understanding the sentiment behind the reviews of an individual and allows an organization to make improvements to its goods and services in response to the feedback. An automatic model is required to classify the sentiments as positive and negative. Thus, the paper proposes a sentence-level sentiment analysis approach to classify whether a movie review is positive or negative. The experiment is performed on the IMDb dataset using three supervised machine algorithms, Linear Support Vector Machine, Logistic Regression and Multinomial Naïve Bayes. Each have been tuned to have the best settings of hyperparameters to achieve the best possible results. The results obtained by these models were then compared based on their Accuracy Score, F1-Score and AUC Score. This approach obtained an F1-Score of 0.914 and an AUC-ROC score of 0.97 after 10-fold Cross-validation. There was an improvement in the evaluation of the F1 score and AUC-ROC score compared to other state-of-the-art models.