Enhancing machine learning-based sentiment analysis through feature extraction techniques.

Area of Science:

• Natural Language Processing
• Machine Learning
• Data Science

Background:

• Feature extraction is vital for sentiment classification performance.
• Selecting optimal feature extraction methods enhances sentiment analysis tasks.
• Machine learning and feature extraction research requires methodical analysis.

Purpose of the Study:

• To analyze and summarize feature extraction techniques from a machine learning perspective.
• To guide the selection of suitable feature extraction methods for sentiment analysis.
• To provide directions for future research in machine learning and feature extraction.

Main Methods:

• Evaluated Bag-of-words (BOW), Word2Vector, N-gram, Term Frequency-Inverse Document Frequency (TF-IDF), Hashing Vectorizer (HV), and Global Vector for Word Representation (GloVe).
• Applied feature extraction techniques to Twitter US airlines and Amazon musical instrument reviews datasets.
• Trained a random forest classifier using 70% training and 30% testing data for performance evaluation.

Main Results:

• Term Frequency-Inverse Document Frequency (TF-IDF) achieved 99% accuracy on the Amazon reviews dataset.
• TF-IDF achieved 96% accuracy on the Twitter US airlines dataset.
• Comparative analysis demonstrated TF-IDF's superior performance over other methods.

Conclusions:

• Feature extraction significantly impacts sentiment analysis model performance.
• TF-IDF is a highly effective feature extraction technique for sentiment analysis.
• The study offers practical insights for improving sentiment analysis models and future research.