Analyzing Sentiments in Social Media Posts through Hybrid Models

Sentiment analysis, a vital natural language processing (NLP) technique, discerns the polarity of textual data-whether it is positive, negative, or neutral. In an era where human expression proliferates online, sentiment analysis stands as a crucial tool, influencing decisions across diverse sectors like E-commerce, healthcare, politics, and online platforms. The challenges in sentiment analysis revolve around accurate data labeling and the development of efficient classification models. Traditional sequential sentiment analysis models face time constraints, prompting the adoption of transformer models for faster, parallelized processing. The proposed hybrid approach in the paper combines automated labeling methods (emoji-based, lexicon-based, and hashtag-based) and a hybrid RoBERTa-BiLSTM model for preprocessing the textual data, resulting in an impressive precision score of 86.67%. Training diverse models on this transformed data results in a classification process with a 70% training dataset, 20% validation dataset, and 10% testing dataset. Furthermore, an ensemble model, incorporating various machine learning and deep learning models with optimized hyperparameters, demonstrates an accuracy of 78.28% in predicting labeled datasets. This innovative hybrid approach demonstrates promise in refining sentiment analysis applications, providing more accurate insights into the sentiments expressed in textual data.