Text and Emoji Based Sentiment Analysis on social media
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This research investigates sentiment analysis on social media during the
newlinepandemic, employing a dual approach that combines traditional Machine Learning
newline(ML) and advanced deep learning techniques. The initial phase involved testing
newlineand fine-tuning common ML algorithms, including Support Vector Machines,
newlineNaive Bayes, Decision Trees, and Gradient Boosting. Fine-tuned Gradient
newlineBoosting emerged as the top-performing model among these algorithm, prompting
newlinefurther exploration into deep learning methodologies. Comparisons between ML
newlineand deep learning models were conducted, incorporating Long Short-Term
newlineMemory (LSTM), Bidirectional Encoder Representations from Transformers
newline(BERT), and a custom model named Covid_SentiBERT. Notably,
newlineCovid_SentiBERT achieved 95.36% of accuracy, outperforming the fine-tuned
newlineGradient Boosting (93.5% accuracy) approach in sentiment analysis on Twitter
newlineduring the pandemic.
newlineAn innovative aspect of this study focused on the analysis of emojis,
newlinerecognizing their pivotal role in expressing sentiment on social media. To address
newlinethis unique challenge, the research introduced a novel approach leveraging
newlineRecurrent Neural Networks (RNN) with LSTM architecture, termed (Fuzzy
newlineButterfly Optimization) FBO-based RNN-LSTM. This specialized model exhibited
newlinevii
newlinebetter performance in Geometric mean 93.2%, Mathews Correlation Coefficient
newline96.8% and F-measure 95.12%, accuracy 95%, precision 93%, recall result 94%
newlinewhich capturing sentiment conveyed through emojis, showcasing its effectiveness
newlinein handling diverse forms of sentiment expression in Twitter data. The findings of
newlinethis research underscore the efficacy of fine-tuned Gradient Boosting in traditional
newlineML approaches, the superiority of Covid_SentiBERT in deep learning
newlineapplications, and the specialized excellence of the FBO - based RNN-LSTM model
newlinein understanding sentiments expressed through emojis.