Machine learning-based opinion extraction approach from movie reviews for sentiment analysis

The field of sentiment mining, also known as sentiment analysis and sentiment extraction, has shown a surge with diverse applications. Various approaches to automate the process of sentiment analysis have been introduced focusing on machine learning, data mining, and natural language processing. Existing works examine the feelings embedded in people’s opinions and beliefs that affect various areas applications, including companies’ services and products. Movie review must undergoes various processes to detect and name feelings and achieve better accuracy in decision making. Due to the structure of the language, the difficulties have been increased for the natural language processing algorithms, particularly for the grammar and dictionary management. In this paper, a new approach for extracting words from a specific text and classifying them into positive or negative opinion classes is proposed. We employ pre-processing and extraction technique for the selection of 500 to 20,000 words based on their frequency, which can be utilized using vector construction. This is performed using “Term Frequency” and “Term Frequency-Inverse Document Frequency”. Four Naive Bayes models (including Complement, Multinomial, Bernoulli, and Gaussian) are implemented and evaluated over unseen samples. Our extensive simulation results indicated that the Naive Bayes multinomial model achieved the best performance with 86.46% accuracy.