A call to action from the White House to help researchers involved in the study of Covid. The startup Indigo.ai and Centro Medico Santagostino launch an artificial intelligence system to identify useful information among 195,000 scientific texts
(AGR) It’s called Record and it’s a Natural Language Processing system that can cut the time needed to find information on Covid-19 for the thousands of researchers who are currently working on it around the world. Less time to find answers, more time to get closer to solutions, especially the vaccine. It was developed by Indigo.ai – the Italian startup specialized in the design and construction of virtual assistants, language technologies and conversational experiences – and the Santagostino Medical Center – a network of complete specialist clinics, the first in Italy to experiment with a model of healthcare that combines high quality and affordable rates.
The cue came from the “CORD-19 – COVID-19 Open Research Dataset Challenge” call launched by the Allen Institute for AI in partnership with Chan Zuckerberg Initiative, Georgetown University’s Center for Security and Emerging Technology, Microsoft Research, IBM, National Library of Medicine – National Institutes of Health and in coordination with the White House’s Office of Science and Technology Policy in order to support research around the Coronavirus and help find a vaccine as soon as possible. The request is the creation of data mining systems that help scholars to quickly update themselves on specific topics of interest, without having to read hundreds of pages every day, within the more than 195 thousand scientific articles on COVID-19, SARS-CoV -2, and other types of Coronavirus.
Record was born from a fruitful synergy, in which Indigo.ai developed the artificial intelligence technology and the Santagostino Center or Doctor supported the learning part of AI by providing a team of experts, to verify that the scientific information identified from the AI were actually correct. To date, RECORD is able to find information relevant to researchers’ requests in 80% of cases.
The Record search engine works in three phases: once a question has been received, through some specific keywords, the system filters the documents that most likely contain the answer; then the subset of the document with a semantic content very similar to the question asked is selected, and finally through a question answering model an answer is provided for each of the selected documents, together with further information such as the title, authors, citations and publication journal with relative scientific impact, to help users evaluate the relevance of the document and the reliability of the response. A search performed through a vectorization of all paragraphs, using the Bert-based sentence embedding model:
A tested method: on the specific questions posed by the experts of the Santagostino Medical Center, the positive answers reached 63%, on the questions extracted from the White House challenge the rate of correct answers was around 80%.
Furthermore, RECORD is a model of NLP that can be replicated, or that can be declined in other contexts: it can be reconverted for business needs or can be reformulated in the event of new emergencies – as it is a system capable of providing any question with the specific answers contained in the documents cataloged as relevant to the object in question . A completely revolutionary approach, because if until now this type of research required human supervision, in the Indigo.ai model everything passes through artificial intelligence.
“Research on Covid-19 is proceeding very fast and consequently the number of papers and publications increases exponentially. In our own small way, we have used our AI and Natural Language Processing skills to make our contribution to the medical-scientific community on the Covid front. We have activated our data scientists and created a model that, we hope, will speed up the process of finding a vaccine and indirectly save lives ”commented Gianluca Maruzzella, Founder & CEO of Indigo.ai.
“We are proud to have participated in this innovative project,” adds Luca Foresti, CEO of the Santagostino Medical Center, “making the experience of our doctors and experts available. For years the Santagostino Medical Center has been pushing a lot on the digitalization of healthcare. And on the occasion of this crisis, for example, he built innovative tools such as video consultations with specialists, video therapies with psychotherapists, a chat for general medicine, he collaborated in the creation of the Immune App. I am convinced that we can get out of the crisis by combining different skills: only in this way will we be able to navigate complexity. “