Detecting Sentiment Polarities with Comparative Analysis of Machine Learning and Deep Learning Algorithms

Abstract

As technology develops, customer reviews are used to assess the quality of products due to the increasing number of selling products online. The extraction of useful minerals from reviews are extremely important for future buyers who are seeking thoughts and sentiments to help their decision-making. Sentiment polarity detection is the process of categorizing the emotions expressed with text, mainly to identify whether the subjectiveness of the writer’s attitude toward the product, or service is positive, neutral or negative. To decrease sentiment mistakes on increasingly complex training data, we deploy machine and hybrid learning models that integrate multiple types of deep neural networks in this study. Then, we apply TF-IDF vectorization for extraction of valuable information in reviews. The paper proposes the comparison of performance between machine learning models, deep learning models and the combination of deep learning models to discover the sentiment polarity on online product reviews. We use the dataset collected from an e-commerce website (Amazon), which includes various product reviews. The experimental results display that combination of deep learning models outperform more machine learning algorithms.