Sentiment Analysis of Nepali COVID-19 Tweets

This repository contains experiments performed using different monolingual and multilingual Transformer models for the Nepali language to perform the sentiment analysis of the Nepali Tweets related to COVID-19. The different monolingual and multilingual models used are as follows which are all available in the HuggingFace.

• amitness/roberta-base-ne
• Sakonii/distilbert-base-nepali
• Rajan/NepaliBERT
• bert-base-multilingual-uncased

DataSets

For this experiment, NepCOV19Tweets is used which contains Nepali tweets related to COVID-19.

Experiments and Results

We experimented with the following hyperparameters:

• Optimizer: AdamW
• Batch Size: 16
• Learning rate: 0.0001

Model Comparsion:

Model	Pre.	Rec.	F1
NepaliBERT	0.31	0.45	0.28
NepBERT	0.70	0.71	0.70
DB-BERT	0.73	0.73	0.73
BERT-bbmu	0.20	0.45	0.28