Artificial Intelligence Based Sentiment Analysis for Social Media

Abstract

Abstract : The exponential growth of social media has established platforms accessed through public APIs as key venues for public discourse and sentiment expression. This research presents a comprehensive AI-based system for real-time sentiment analysis of social-media streams using advanced transformer-based language models. To address challenges related to noisy, short, and multilingual the system integrates a robust preprocessing pipeline capable of handling linguistic variability, and informal writing patterns. The primary objective is to develop an end-to-end sentiment-classification platform that not only performs real-time inference but also enables users to experiment with, evaluate, and fine-tune multiple sentiment-analysis models through an intuitive graphical interface. Rather than focusing on isolated datasets, the methodology leverages diverse large-scale social-media sources and multilingual input streams to ensure realistic, generalisable performance. The system architecture incorporates multiple model categories: traditional deep-learning models such as LSTM and CNN for lightweight inference, and transformer-based models including RoBERTa and XLM-ROBERTa for enhanced multilingual classification. A unified model management layer enables seamless switching between locally trained models and pretrained transformer models across the application. The resulting platform supports real-time sentiment monitoring, multilingual analysis, model training and evaluation, transformer fine-tuning, and interactive visualizations, providing an accessible research environment suitable for users.