Epidemic Question Answering: question generation and entailment for Answer Nugget discovery.

Area of Science:

• Information Science
• Medical Informatics
• Natural Language Processing

Background:

• The COVID-19 pandemic generated a massive volume of information, making it difficult for individuals to find accurate answers.
• Information seekers faced challenges in obtaining specific and up-to-date details regarding COVID-19, SARS-CoV-2, and public health recommendations.

Purpose of the Study:

• To design and evaluate a Question Answering (QA) system for ad-hoc queries related to COVID-19 and SARS-CoV-2.
• To assist both health experts and general consumers in navigating the complex information landscape during the pandemic.

Main Methods:

• Developed a QA system integrating relevance models with automatic question generation from relevant text.
• Utilized question entailment for answer pinpointing and prioritizing novel results.
• Employed state-of-the-art models for question generation and entailment.

Main Results:

• The QA system achieved state-of-the-art performance for questions posed by experts (e.g., researchers, clinicians).
• Competitive results were obtained for questions from health information consumers.
• Over 50% of answers were missed, primarily due to limitations in the employed relevance models.

Conclusions:

• The QA system demonstrates strong performance for expert queries but requires enhancements for consumer queries.
• Future development should focus on improving relevance models and incorporating advanced inference capabilities.
• Accounting for the distinct answer expectations of experts and consumers is crucial for future QA system design.