Automatic question answering for multiple stakeholders, the epidemic question answering dataset.

Area of Science:

• Information Science
• Computational Linguistics
• Public Health

Background:

• The COVID-19 pandemic resulted in an exponential increase in publications covering health, socio-economic, and cultural impacts.
• Manual management of this vast information stream is impractical for clinicians, researchers, policymakers, and the public.
• Automatic Question Answering (AutoQA) offers a solution for efficiently extracting salient information.

Purpose of the Study:

• To develop a dataset for exploring AutoQA systems tailored to diverse stakeholder needs during the pandemic.
• To analyze the information-seeking behaviors and answer generation requirements of various user groups.
• To assess the potential of AutoQA for handling complex information domains beyond public health emergencies.

Main Methods:

• Compiled a comprehensive dataset including scientific articles, government reports, news, social media, and user-generated questions.
• Collected questions from researchers, clinicians, and the general public to represent different stakeholder groups.
• Analyzed the overlap and divergence in information needs and optimal answer sources across stakeholder categories.

Main Results:

• Information needs of experts and the public often overlap but require different sources or answer generation strategies.
• Satisfactory answers for different stakeholders may necessitate distinct information retrieval or synthesis approaches.
• The developed dataset facilitates research into stakeholder-specific AutoQA for pandemic-related information.

Conclusions:

• Automatic Question Answering systems can effectively manage and disseminate critical information during health crises.
• The dataset supports the development of adaptable AutoQA solutions for diverse user groups and information domains.
• This approach holds promise for improving information access in complex fields like law and finance.