Identifying COVID-19 english informative tweets using limited labelled data

Area of Science:

• Natural Language Processing
• Public Health Informatics
• Machine Learning

Background:

• Identifying informative tweets is crucial for real-time COVID-19 monitoring.
• Current methods require extensive labeled data, which is costly and time-consuming.
• A need exists for efficient approaches using limited labeled data.

Purpose of the Study:

• To develop a data-efficient method for identifying informative COVID-19 tweets.
• To achieve state-of-the-art performance with minimal labeled data.
• To address the limitations of existing supervised learning approaches.

Main Methods:

• Utilized a labeled data-efficient approach starting with a small labeled dataset.
• Employed data augmentation techniques to expand the training set.
• Fine-tuned a model using the augmented dataset for informative tweet identification.

Main Results:

• Achieved a state-of-the-art F1-score of 91.23 on the WNUT COVID-19 dataset.
• Demonstrated high performance using only 1000 labeled tweets (14.3% of the full set).
• Established a new benchmark for identifying COVID-19 informative tweets with limited data.

Conclusions:

• The proposed labeled data-efficient approach is effective for identifying informative COVID-19 tweets.
• This method significantly reduces the reliance on large labeled datasets.
• It offers a practical solution for building timely public health surveillance systems.