Sentiment Analysis of Code-Mixed WhatsApp Conversations (Hinglish): A Hybrid Approach

This paper presents a framework for sentiment analysis of WhatsApp chat conversations, focusing on classifying the text messages as positive, negative, or neutral. As WhatsApp continues to gain prominence as a global communication tool, studying the sentiments expressed in everyday, often multilingual languages, specifically Hinglish, can help understand social relations, general attitudes or even the mental well-being of the communities. In the proposed framework, data pre-processing is done via the Natural Language Processing (NLP) technique with text cleaning, tokenization and stopword removal (including removal of English and custom hinglish words). The VADER lexicon is used for sentiment labeling, and then an LSTM-based deep learning model is trained with an accuracy of 97.89%. Visualizations using Matplotlib and Seaborn display sentiment trends through distribution bar graphs, line graphs, and word clouds to enhance interpretability. The entire procedure is functionalized into a Streamlit-Based web app where users can input and analyze their WhatsApp chat data.