Twitter Sentiment Analysis Full-Stack Deployment

Suyash Gawande¹ *,

• 1, ,

Social media platforms like Twitter provide
a continuous stream of public opinion in the form of
vast amounts of informal text, which presents challenges
for automated analysis. This paper outlines a complete
end-to-end system for performing sentiment analysis on
Twitter data. The system utilizes robust preprocessing
techniques, an LSTM-based deep learning classifier, and is
deployed with a full-stack architecture featuring a Python
backend and a React frontend. We focus on practical
aspects such as dataset curation, tokenization, model train-
ing, and evaluation. The methodology is detailed, along
with an experimental evaluation that includes an ablation
study comparing LSTM with CNN and transformer-based
models. A user study was also conducted to assess the
system’s usability. The paper acknowledges limitations like
handling sarcasm and domain bias and suggests future
enhancements, including transformer model fine-tuning
and multilingual support.