DeepCovid: An NLP System for Analyzing COVID-19 News Articles
Author Information
Author(s): Liu Hongfang, Zheng Jiaping, Gan Wenyi, Ravaut Mathieu, Zhao Ruochen, Phung Duy, Qin Vicky Mengqi, Milovanovic Dusan, Pienkowska Anita, Bojic Iva, Car Josip, Joty Shafiq
Primary Institution: Nanyang Technological University
Hypothesis
Can natural language processing (NLP) be used to efficiently analyze a large volume of COVID-19 related news articles to support health workforce policy decisions?
Conclusion
The DeepCovid system effectively combines NLP models with human analysis to provide timely insights into the impacts of COVID-19 on health workers.
Supporting Evidence
- DeepCovid was trained on 2.8 million news articles from over 3000 sources.
- The system achieved a binary classification performance of 98.98 for ROC-AUC.
- DeepCovid's extractive summarization component had a mean ROUGE score of 47.76.
- Human experts used DeepCovid's summaries to write reports on the COVID-19 pandemic.
Takeaway
DeepCovid is a smart computer program that helps people quickly understand a lot of news about COVID-19 and how it affects health workers.
Methodology
The study used a combination of rules-based classification and deep learning models to analyze 2.8 million news articles from the WHO EIOS database.
Potential Biases
Potential bias from focusing only on English news sources may overlook important information from non-English reports.
Limitations
The study was limited to English-language articles and required human annotations for model training, which can be time-consuming.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website