An Optimized Machine Learning Approach for Scalable Online Public Opinion Mining

Abstract

The swift growth of the social media platforms resulted in tremendous amounts of unstructured textual data which means that online analysis of public opinion is a crucial activity in comprehending the perception and sentiment dynamics in the society. A good sentiment classification needs a good pre-processing, feature selection methods optimized to deal with noise, high dimensionality and scalability. This research introduces a unified model of online sentiment analysis of public opinion about something based on the Contextual-Adaptive Text Refinement Engine (CATR-E) to do text pre-processing, Federated Feature Ranking Protocol (FeFRP) to determine features, and Logistic Regression to classify sentiment. CATR is a systematic method of noise elimination emojis, URLs, and stop words and semantic relevance normalization and lemmatization. FeFRP identifies and chooses the most informative TF-IDF features which tremendously downsizes dimensionality, yet still captures important sentiment-carrying words. Logistic Regression is subsequently used to determine sentiments into predetermined categories according to the estimation of probabilities. The framework is effective as seen through experimental assessment on a real-world Twitter entity sentiment dataset with a Feature Selection 93.00 better than some of the current machine learning and deep learning Existing methods. The effectiveness of the framework is supported by its 93.00% accuracy of feature selection and 35.04% feature dimensionality reduction using the FeFRP protocol as well as 43.79% of token reduction using the CATR-E engine that is far better than the current machine learning and deep learning methods. The findings confirm that the joint pre-processing and feature selection measure is improved to classify the data and is still interpretable and computationally efficient. In general, the framework enables a dependable and scalable approach to large-scale, real-time analysis of the public opinion on social media.