Computing Reviews

Identifying informative tweets during a pandemic via a topic-aware neural language model
Gao W., Li L., Tao X., Zhou J., Tao J. World Wide Web2655-70,2023.Type:Article
Date Reviewed: 04/28/23

Years ago, when Twitter was a new social media tool, I reviewed an article for Computing Reviews that looked at how tweets can be used in emergency situations. This new paper follows along similar lines: Twitter can be used to help with specific emergency situations, for example, in the case of a pandemic like Covid. It can help emergency response organizations to prioritize tasks and make better decisions.

However, the same issues uncovered then still prevail today, especially because of the shortness of tweets and the difficulty of determining their relatability, thus making most of them non-informative. The authors propose a modification to the existing Bidirectional Encoder Representation from Transformers (BERT) model with the use of what they call topic-aware BERT (TABERT). They hope to reduce the amount of rich data that is classified into wrong categories by considering not only location, content, and error message propagation, but also topic information and unlabeled data that is often ignored. In their experimentation, by removing tweets of less than five words, emojis, and small letters, to name just a few, they were able to connect unlabeled data in short communications to find actionable threads. The proper response agency can then do prevention and control work in the form of caution and advise, knowing emerging threats, rendering services, sending volunteers, and maybe also identifying new threats.

The authors use mathematical equations and graphs to show their work and results. The paper is riddled with acronyms, which may require a cheatsheet to make heads or tails of some of the material as one reads along.

At a time when there is still controversy about the origins of Covid and how to conquer it, it is a good idea to use tweets as a way to find information. Neither I nor my students anticipated that incorporating Twitter into my classes would prove to become a useful tool in many situations. In this case, the proposed TABERT model to detect informative posts on social media platforms serves as a foundation for conversations that can help governments, scientists, policy-makers, and individuals to make informed decisions regarding this pandemic.

Reviewer:  Cecilia G. Manrique Review #: CR147583 (2307-0096)

Reproduction in whole or in part without permission is prohibited.   Copyright 2024 ComputingReviews.com™
Terms of Use
| Privacy Policy