|
Activity Number:
|
74
- Text Analysis in Machine Learning and Statistical Models
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Section on Statistics in Defense and National Security
|
|
Abstract #312997
|
|
|
Title:
|
Dynamically Evolving Transformer Models for Article Tagging for Biosurveillance
|
|
Author(s):
|
Karl Pazdernik* and Samuel Dixon and Daniel Farber and Aaron Tuor and Andrew Barker and Elise Saxon and Lauren Charles
|
|
Companies:
|
Pacific Northwest National Laboratory and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory and Pacific Northwest National Laboratory
|
|
Keywords:
|
Natural Language Processing;
Transformers;
Dynamic Modeling;
Biosurveillance;
COVID-19;
Text Analysis
|
|
Abstract:
|
Automated tagging of text is a useful technique that can streamline many online services. However, reliable ground-truth labels can be difficult or impossible to attain especially in enough quantity to properly train supervised learning methods. Even the concept of “enough data” in text analytics is not well-understood. We explore these problems in an application of biosurveillance using news articles where the discovery of new event characteristics, such as the collapse of a water and wastewater system, a novel disease-causing agent, or a border closure, is critical to global monitoring of disease. We provide specific examples of how the COVID-19 outbreak spread across the globe and evolved over time. We use evolving transformer models to dynamically update predictions as more training data (manually tagged articles) becomes available.
|
Authors who are presenting talks have a * after their name.
