Classification of movie reviews using term frequency-inverse document frequency and optimized machine learning...

Area of Science:

• Natural Language Processing
• Machine Learning
• Data Science

Background:

• The Internet Movie Database (IMDb) hosts millions of user reviews, offering rich data for sentiment analysis.
• Manual analysis of vast review datasets is impractical, necessitating automated sentiment analysis tools.

Purpose of the Study:

• To implement and evaluate machine learning models for sentiment polarity detection in IMDb user reviews.
• To identify the optimal combination of feature engineering techniques and classification models for accurate sentiment analysis.

Main Methods:

• Preprocessing of IMDb reviews to remove noise and redundant information.
• Application of feature engineering techniques including Term Frequency-Inverse Document Frequency (TF-IDF), Bag of Words, Global Vectors for Word Representations (GloVe), and Word2Vec.
• Implementation and hyperparameter tuning of classification models such as Support Vector Machines (SVM), Naïve Bayes, Random Forest, and Gradient Boosting.
• Utilizing TextBlob for sentiment assignment to address contradictions in user-assigned labels.

Main Results:

• Support Vector Machines (SVM) combined with TF-IDF features achieved an accuracy of 89.55%.
• The use of TextBlob for sentiment assignment improved the overall sentiment classification accuracy to 92%.

Conclusions:

• Machine learning models, particularly SVM with TF-IDF, are effective for sentiment analysis of online movie reviews.
• TextBlob's sentiment assignment pre-processing step can significantly enhance the accuracy of sentiment classification models by mitigating label inconsistencies.