Multilingual Sentiment Analysis

A multilingual analysis system designed for sentiment classification across four different languages: English, Hindi, Marathi, and Gujarati.

Setup

• Input: Any sentence in English, Hindi, Marathi, and Gujarati.
• Output: Sentiment Class (POSITIVE or NEGATIVE)
• The system features an intuitive GUI for seamless web interface interaction.

Training

The model has undergone fine-tuning on target language data, facilitated by the Google Translation API.

Dataset: Sentiment140

Two distinct model architectures have been evaluated:

1. DistilBERT
2. mBERT

Results

Model	Class	Precision	Recall	F1-Score
DistilBERT	0	0.77	0.82	0.80
	1	0.81	0.76	0.78
mBERT	0	0.71	0.74	0.73
	1	0.74	0.71	0.72