Code-Mix Sentiment Analysis on Hinglish Tweets for Brand Monitoring in India Using Transformer Models

This paper explores sentiment analysis of Hinglish tweets with the aim of supporting effective brand monitoring in India’s multilingual social media environment. By adapting transformer models such as mBERT and XLM-R to code-mixed data and utilizing refined subword tokenization, the study addresses irregularities common in Hinglish communication. Model selection was motivated by the need for a robust approach capable of capturing both lexical variation and informal expressions found in social platforms. Evaluation using curated Hinglish sentiment datasets demonstrates that transfo.rmer-based architectures reliably outperform standard baselines, yielding improved accuracy and greater resilience to language mixing. The results highlight the practical utility of multilingual NLP for real time assessment of consumer sentiment, providing valuable guidance for brands seeking to navigate the dynamics of India’s online discourse.