There are lots of COVID-19 literature out there. How can we quickly give researchers the insights they need to support the fight against this coronavirus pandemic?
What it does
We uploaded a set of bioRxiv papers about COVID-19 to IBM Watson. Then we created a cognitive search engine. Users can send natural language queries along with metadata such as concepts and entities that were automatically extracted from the papers.
Challenges we ran into
Lack of domain expertise in virology, epidemiology, and medical research.
Accomplishments that we're proud of
With this tool, users can quickly zero in on the few most relevant articles from a collection with millions of documents. No need to trawl through pages of search results. The tool highlights metadata such as concepts, entities, and keywords, allowing users to refine search queries and get the gist of paragraph sections without reading through them.
What we learned
This virus spreads rapidly so we wanted to build something that can be up and running quickly. No time to tinker with source code libraries like TensorFlow and make things from scratch. Speed is of the essence so we built on top of Watson's artificial intelligence platform.
What's next for Biosight ☣️
- Work with subject matter experts and researchers to figure out research questions and data sources that may contain the answers.
- Reduce this web form to a chatbot, so users can just enter natural language questions without having to know about technical details like concepts and entities.
- Create a custom machine learning model for infectious diseases.