Machine Learning: Contextual Sentiment Understanding Through Hybrid Computational Intelligence Models

Abstract: As a consequence of the escalating amounts of user-generated text available today, including tweets, comments, reviews, and blog posts, researchers can examine and derive knowledge about public opinion. Sentiment analysis, a pivotal application area of Natural Language Processing (NLP), interprets and categorizes opinions to assess emotions and attitudes that are expressed in text. This study presents a review of former studies regarding prediction accuracy using sentiment analysis (with a focus on supervised machines, deep learning, and combining of approaches). For example, we analyze various supervised algorithms including Support Vector Machines, Logistic Regression, K-Nearest Neighbor, Naïve Bayes, and Random Forest, and introduce advanced neural models including Recurrent Neural Networks and Long Short-Term Memory networks. Also, some of our research describes new hybrid architectures that combine keyword extraction algorithm based on graph methods and machine learning models that improve performance on contextual text using the first applications of statistical models of sentiment for real datasets. We also examine data preprocessing and feature selection efforts, as well as the use of ensemble models for classification performance. Our review finds that hybrid models and those using graph-based techniques may be advantageous compared to standard supervised models, as well scalable, adaptive models for sentiment mining, brand communication monitoring, and decision support.

Keywords: Sentiment Analysis, Natural Language Processing (NLP), Machine Learning, Deep Learning, Hybrid Models, Support Vector Machine (SVM), Recurrent Neural Network (RNN), LSTM, Graph-Based Approach, Text Classification, Opinion Mining.