Sentiment Analysis for COVID-19 Related Tweets Using Deep Bi-Directional Transformers
2022 IEEE Region 10 International Conference, TENCON 2022
; 2022-November, 2022.
Article
in English
| Scopus | ID: covidwho-2192086
ABSTRACT
This study uses a pre-trained Bi-directional Encoder Representations from Transformers (BERT) with an AdamW optimizer for sentiment analysis of COVID-19 related tweets. This is performed on around 32,000 tweets from an annotated dataset of 190 million tweets. The sentiment of each tweet was predicted between three different classes, negative, neutral, and positive. Under sampling was performed to balance out the data and the model was fine-tuned over 4 epochs. The resulting model was best at predicting negative sentiment and worst at predicting neutral sentiment. The resulting accuracy was found to be 75.15%, however, increasing the amount of data used would likely improve this significantly. © 2022 IEEE.
Full text:
Available
Collection:
Databases of international organizations
Database:
Scopus
Language:
English
Journal:
2022 IEEE Region 10 International Conference, TENCON 2022
Year:
2022
Document Type:
Article
Similar
MEDLINE
...
LILACS
LIS