Sentiment Analysis of Social Media Using Natural Language Processing (NLP)

Priyam Prashant Pal

Abstract: Social media platforms- think Twitter/X, Reddit, Facebook, Instagram- churn out billions of posts and comments every day. It's a nonstop stream of real-time thoughts, feelings, trends, and cultural shifts. For anyone interested in what people really think or feel, that's a goldmine. Trouble is, there's just so much data, and it comes in all shapes and sizes, so making sense of it isn't simple. That's where sentiment analysis comes in. Basically, it's about teaching computers to spot emotions, opinions, and attitudes in all that text- like telling if someone's happy, annoyed, or just neutral. In the early days, this mostly meant looking for keywords. But thanks to big strides in Natural Language Processing (NLP), and the rise of huge transformer models, sentiment analysis can now catch context, get sarcasm, or even understand when someone's talking about one aspect of a product but not another. This paper reviews how researchers make sense of social media sentiments, from start to finish. We break down everything- how we gather the data, clean it up, pull out key info, train models, and judge how well they do. We put classic machine learning methods like Naive Bayes and SVM side-by-side with deep learning setups like LSTM and CNN, and then with cutting-edge transformers like BERT, RoBERTa, and BERTweet. Here's the punchline: transformers leave the older techniques in the dust, especially those models tweaked for social media-they hit up to 95% accuracy on real datasets. We also touch on how this is being used in the wild right now, the problems that still need solving, and what's coming next.

Keywords: sentiment analysis, natural language processing, social media, BERT, BERTweet, deep learning, opinion mining, text classification, transformer models