Element AI’s search tool surfaces curated coronavirus studies
Element AI today released a search tool that combs through the COVID-19 Open Research Dataset, a repository of over 44,000 scholarly articles about COVID-19 and related coronaviruses, for papers potentially of use to researchers. Users can search or query natural language terms, phrases, and keywords to surface articles that contain semantically similar content, or copy paragraphs of text or questions into the search bar to return articles with only the most important sentences highlighted.
A deluge of studies on the novel coronavirus, which is projected to sicken millions of people, have made their way to the web in the months since the outbreak began. (According to Reuters, at least 153 preprint studies about COVID-19 have been made publicly available as of March 24.) They promise insights into the virus’ spread, but many haven’t been peer-reviewed, making it difficult for stakeholders to sort the wheat from the chaff.
To this end, Element AI’s tool leverages tech from the company’s Knowledge Scout product, which uses AI to capture the relationships among different pieces of information, to learn and improve over time while building a repository of tacit knowledge. Element AI says that the platform will be progressively updated in the coming weeks with additional COVID-19 data sets, alongside features including open-domain question-answering capabilities, query-driven summarization, and topic discovery.
The launch of Element AI’s platform follows that of Vespa’s CORD-19 Search, which similarly trawls the COVID-19 Open Research Dataset for vetted research papers. For its part, Korea University’s DMIS Lab this month released Covidsearch, which provides real-time question-answering on 31,000 COVID-19-related articles with results that highlight relevant biomedical entities. And the Allen Institute for AI offers a no-frills platform that searches the full text of the COVID-19 Open Research Dataset.
The AI underpinning these and other COVID-19 search tools learn from signals (i.e., data derived from various inputs). Each signal informs the system’s predictions such that it learns how various resources are relevant (or not) to a search query. Natural language processing enables them to understand a piece of research in the context of a data set, while natural language search — a specialized application of AI that creates a “word mesh” from free-flowing text, akin to a knowledge graph — connects similar concepts that are related to larger ideas to return the same answer regardless of how a query is phrased.
The jury is out on how big of an impact semantic search tools might have on continuing COVID-19 research, but as alluded to earlier, they might tamp down on the more questionable research that has come to light. One recent paper suggests a link between the new coronavirus and HIV, while another claims it’s from outer space.