The Allen Institute for Artificial Intelligence has added new features to its academic search engine, Semantic Scholar, to make it easier for professionals and plebs to understand and advance research.
Funded by the late Microsoft cofounder Paul Allen and his sister Jody Allen, an entrepreneur and philanthropist, the Allen Institute for Artificial Intelligence (AI2) is a nonprofit research institute focused on building AI systems that can understand science.
Semantic Scholar can’t quite do that, but people can - with the right information - and Semantic Scholar might help them. It’s a search engine that pulls the most relevant information for users looking up scientific papers. Built three years ago, the tool is used by two million people each month.
Here’s a link to an example of how it works for a machine learning paper titled: “Why Should I Trust You?": Explaining the Predictions of Any Classifier”.
The whole search engine is built from a knowledge base. Researchers and engineers built tools to like Science Parse, to automatically extract metadata from the PDF files of papers to take the titles, author information, abstract, and references and adds it onto Semantic Scholar. DeepFigures takes all the useful graphs and tables in the paper.
The new features include a column on the right hand side charting the paper’s impact. Natural language processing is used to search for keywords linked to the article to see how often it has been cited by other researchers or discussed on mentioned in Twitter discussions.
"Male researchers are about 33 per cent more likely than female researchers to cite themselves in research papers. We remove self-citations to level the playing field so that the metrics are better," Oren Etzioni, CEO of AI2, explained to The Register.
What’s also handy is all the stuff that’s collected under “Supplemental Content”. The knowledge base crawls the internet looking for any pages that mention the paper, and decides if the information is useful to add, Doug Raymond, general manager of the Semantic Scholar project, explained to The Register.
Finally, a use for AI and good old-fashioned simulations: Hunting down E.T. in outer spaceREAD MORE
This can include YouTube videos that explain the research or recorded talks where the researchers have presented their work at conferences. It also has links to the GitHub repo if the associated code is public, PowerPoint presentations and also a series of links to blogs and news articles discussing the work.
All of this done automatically for over 40 million computer science and biomedicine papers scraped from Arxiv, and journals like Nature or PubMed.
“It’s for professors or educated consumers, anyone who wants to understand what’s going on in science. Semantic Scholar helps cut through “information overload”, and help answer the question: Should I read this paper?,” Etzioni said.
“It lets people from different backgrounds really dive into a subject. They can look at key results, whether its from clinical trials or data in graphs or tables. They can look at the software associated with it in an easy layout, and this could help drive research forward.”
At the moment, it takes a while for new papers to pop up on the search engine, up to two weeks for Arxiv ones. “We’re in the process of updating the system so the wait time goes down to about 48 hours,” Etzioni added. AI2 also hopes to expand to include other subjects like biology, physics or chemistry. ®