A better way to search through scientific papers

ARTIFICIAL intelligence (AI) is not just for playing games. It also has important practical uses. One such is in Semantic Scholar, a system developed by researchers at the Allen Institute for Artificial Intelligence, in Seattle, for the purpose of ferreting out the scientific papers most relevant to a particular problem. This week Marie Hagman, the project’s leader, and her colleagues have launched an updated version of the system. They have added 26m biomedical-research papers to the 12m previously contained in its database, and upgraded the way that the database’s contents can be searched and correlated. Instead of relying on citations in other papers, or the frequency of recurring phrases to rank the relevance of papers, as it once did and rivals such as Google Scholar still do, the new version of Semantic Scholar applies AI to try to understand the context of those phrases, and thus achieve better results.

Like most AI systems, the new Semantic Scholar relies on a neural network—a computer architecture inspired by the way real neurons connect to each other. Neural networks are able to learn tasks by trial-and-error. Ms Hagman’s team wished to bend their network to the task of recognising scientific phrases and their contexts.