Practical AI COVID-19 Q&A and CORD-19
Apr 6, 2020
Timo Möeller, co-founder of Deepset AI, and Tony Reina, chief AI architect for health at Intel, discuss their innovative COVID-QA system designed to answer pandemic-related questions. They delve into the integration of the CORD-19 dataset, emphasizing the importance of expert input and community involvement. The duo highlights the challenges of misinformation and the need for accurate data to support public health, all while showcasing AI's potential to enhance accessibility and collaboration during these critical times.
AI Snips
Chapters
Transcript
Episode notes
COVID-QA Origins
- COVID-QA arose from a German government hackathon focused on factual information access.
- 25 developers rapidly built the UI, backend, and scrapers during the event.
Information Needs
- People need reliable COVID-19 information, especially concerning its spread and prevention.
- This is crucial in densely populated areas where misinformation can easily proliferate.
Technical Approach
- COVID-QA uses Elasticsearch for basic question matching and BERT embeddings for more advanced semantic matching.
- Sentence transformers enhance BERT's performance by creating Siamese networks trained on user questions and FAQ data.