Turn the stone into gold!

AI successfully unearthed the paper "treasure"

  Helps rationally allocate scientific research resources in a data-driven way

  Science and Technology Daily, Beijing, May 18 (Reporter Zhang Mengran) British "Nature Biotechnology" magazine published a research that spans artificial intelligence and biotechnology on the 18th. A machine learning model developed by the Massachusetts Institute of Technology team can be used for prediction. The future impact of published research in the scientific literature.

At present, the scoring of this model can be used to predict the "top 5% of papers" published in any year, which will complement the current bibliometric analysis system that relies on paper citation indicators.

  At this stage, there are many systems that are used to evaluate the research output of researchers, including indicators based on the citations of their papers.

With the rise of machine learning in the field of artificial intelligence, scientists believe that researchers can judge the potential impact of their published research from more perspectives on the output of researchers.

  To this end, the Massachusetts Institute of Technology research team launched a machine learning model that can predict web page ranking scores on a time scale—similar to the index used to rank web pages for importance, and proposed that the model be used for evaluation The output of the researcher.

  In order to realize this idea, scientists James Weiss and Joseph Jakobson established a model called "Dynamic Early Warning by Learning and Predicting High Impact", which was trained with scientific research graphs.

The data set used by the research team contains 16,878,850 unique papers published between 1980 and 2019, from which 29 characteristics related to each paper, author, journal, and network from 1 to 5 years after the paper was published .

The research team then used the characteristics of each paper to train a machine learning model, and let the model give an impact "early warning" score.

  In a retrospective blinding study, this latest model accurately identified 19 of the 20 biotechnologies with significant impact between 1980 and 2014.

This model also predicts that 50 papers published in 42 journals in the field of biotechnology in 2018 may be among the top 5% of the future rankings. This result will be able to be discovered in a data-driven manner and allow funding to flow to those "hidden" "Good research.

  The researchers said that before applying this type of model to other research areas, further testing is needed to compare the method's performance in areas other than biotechnology with conventional impact indexes, such as field-normalized citation scores.