A Sampling-Based Sentiment Analysis of Imbalanced Streamed Movie Reviews

Sentiment analysis has gained significant importance in analyzing individuals' attitudes and perceptions toward various products, services, and entertainment mediums, including movies. Evaluating the sentiment expressed in movie reviews can provide valuable insights into how users interpret and react to specific films. However, Movie review datasets often suffer from an imbalance in the distribution of positive and negative sentiment labels, which presents challenges for accurate sentiment classification. We propose a framework that harnesses streaming data for enhancing sentiment analysis algorithms. First, we create an initial model using an IMDB movie review dataset to categorize real-time review streams. To address the issue of imbalanced streamed data in movie reviews, we apply diverse sampling techniques, mitigating bias toward the dominant sentiment. This method bolsters the sentiment classifier's effectiveness. Additionally, we iteratively improve the initial model using recorded classification outcomes. We conducted comprehensive experiments on varied movie review datasets to assess our approach's effectiveness. Evaluation metrics were used for comparison, including accuracy, precision, recall, and F1-score. The results encompassed contrasting our sampling-driven method with baseline approaches. The SVC outperformed other algorithms in a native classification environment, whereas the extra tree excelled in a streamed classification environment. These outcomes underscored our framework's efficacy in enhancing sentiment analysis algorithm performance.