Semantic Scholar ranked the most influential neuroscientists

If we talk about research that influenced the development of neuroscience, the University College London (UCL) has something to boast. This conclusion is not the opinion of an expert or the result of the analysis of the vast staff of some analytical agency: all the work was done by a computer.

The Semantic Scholar program analyzed the content of 2.5 million scientific articles and the citation of their authors, and then calculated the assessment of the influence of each author on the rest. As a result, it turned out that three of the most influential scientists in this field work for the benefit of science at University College London: Karl Friston, a specialist in parametric methods of statistics (1st place), Raymond Dolan, expert in the field of emotional influence on cognition (2nd place) and Chris Frith, a researcher in the field of cognitive schizophrenia and social cognition (7th place).
')

Semantic Scholar is an online tool that was created in the laboratory of the Allen Institute of Artificial Intelligence (AI2) in Seattle, Washington. The April debut of Semantic Scholar made a pleasant impression: the service was rated by the most influential scientists in the field of computer science, based on 2 million academic papers. Since then, the AI2 team has expanded its article base to 10 million, a quarter of which is work in the field of neurobiology. The team of scientists is going to increase the base of biomedical literature to 20 million documents by next year.

When Semantic Scholar analyzes an article, he sees more than a typical academic search engine and much more than a person. According to project manager Oren Etzioni, AI2 CEO, his team used machine learning, processing of natural language and computer vision technology in their work in order to delve into the semantics.

In order to assess the possibilities of the Semantic Scholar, scientists propose to look at the results of a semantic analysis of scientific works, in which the basal nuclei of the songbirds are examined from different angles. In the left area of ​​the screen, we see the keywords that the service identified from these documents: not only traditional bibliographic data, such as the date of publication and information about the authors, but also the cell types used in the experiments, and even methods.



The development of researchers from AI2 interested and other scientists. So, Sam Gershman, a specialist in computational neuroscience from Harvard University , tried out Semantic Scholar. He said that this is a very interesting tool with undeniable advantages over Google Scholar. For example, in Semantic Scholar there is the possibility of more fine-tuning the sorting of articles. In addition, it shows links to the article, some numbers and diagrams.

At the same time, Gershman immediately discovered a problem that affects all search engines: poor quality data, or “dirty data”. In some works the names of the authors do not match. Yes, and the ambiguity of some terms inhibits the work of the search engine. In addition, the metadata of the research also contains errors: one of Gershman’s works dates from 1987, when the scientist was only two years old.

The most mysterious in this whole story for Gershman was the fact that the articles published in the most influential publications do not get high points: “None of the most influential articles by Thomas Griffis from the University of California at Berkeley is in the top five most cited articles. It's strange. ”Says Gershman.

Oren Etzioni stresses that work on the Semantic Scholar continues. He admits that the service is not perfect and may give errors. Despite this, the tool quite successfully coped with the ranking of the most influential neuroscientists based on current data. It turned out that three of them knew each other from the very beginning of their careers. “We have been working at UCL since 1993 in the same department,” notes Chris Frith. He also added that the Semantic Scholar worked quite correctly.

Topping the list, Karl Friston is the first developer of methods for analyzing visual data of the brain and the creator of a computer model of the brain. When he was informed that he was the first in the TOP 10 scientists, he took this news with a certain amount of humor: “My first thought was“ To whom can I say this and not seem indiscreet? ”. Then I realized that the only people who would like to hear about this are my children! ”

The need to create a service capable of assessing the contribution of a scientist to the development of science arose quite a long time ago. The main difficulty on this path lies in the fact that the influence of the researcher is difficult to measure. So far, the citation index has helped to cope with this task, but in the end, such a counter became the cornerstone of the metrics of the academic publishing industry. Not all quotes can be considered equivalent. Agree that the inspired quoting of whole pages of work differs from a brief mention of the name of the work in the list of sources used. That is why the scientific environment needs a tool that can conduct semantic analysis and produce a more accurate result.

In the future, scientists from the Allen Institute for Artificial Intelligence plan to develop their Semantic Scholar project and turn it into “Siri for Science”. The main goal is for the system to learn to recognize questions in English and search for answers to them.

The work was published in the journal ScienceInsider November 11, 2016
DOI: 10.1126 / science.aal0371