Sentiment Classification of Movie Reviews Based on the Ensemble Machine Learning Model
Zicheng Gan · Advances in computer science research · 2023
Film reviews play a pivotal role in influencing audience decisions, necessitating accurate classification of sentiments as positive or negative, which holds significant importance for the film industry.To address this, the present study introduces an innovative ensemble learning approach that integrates artificial neural networks, LightGBM, and logistic regression models through a stacking technique.The ensemble model is empirically examined using the IMDB dataset, with a comparative analysis conducted against an individual Artificial Neural Network (ANN) model.The findings demonstrate remarkable enhancements, particularly in terms of accuracy and other relevant metrics, achieved by the ensemble model compared to the individual ANN model, specifically yielding an increase in accuracy from 0.8791 to 0.8904.This substantiates the substantial improvement in accuracy offered by the ensemble model, thereby underscoring the efficacy and potential of ensemble learning for sentiment classification in movie reviews.Moreover, an analysis of the confusion matrix reveals that the ensemble model predominantly improves the classification of reviews labeled as 'positive,' as evidenced by an increase in true positive instances from 4359 to 4451, accompanied by a decrease in false positive instances from 668 to 576.By amalgamating predictions from distinct models, the ensemble model effectively mitigates the limitations inherent to individual models and attains superior performance compared to relying on a single model alone.