Online tool breaks down physicists’ research interests
Terms like cosmologist, string theorist, or particle physicist rarely capture the true scope of a scientist’s work. A new website aims to provide a more complete picture. Launched in July, Scimeter allows researchers to create word clouds based on the topics of their arXiv papers, search for scientists with similar interests, and find experts whose work best matches given keywords.
By more effectively disseminating the research interests of those in the physical science community, the site’s creators hope to remove some of the biases in recruiting conference speakers. Rather than relying on networking or other subjective approaches, a conference organizer can seek out researchers whose work best fits a desired topic, says Scimeter’s Sabine Hossenfelder, a theoretical physicist at the Frankfurt Institute for Advanced Studies in Germany. Scimeter could also prove useful when searching for relevant literature or for reviewers qualified to critique a particular paper.
The free-to-use site, which is funded by the nonprofit Foundational Questions Institute, matches researchers and their topics of interest by scanning the titles and abstracts of arXiv papers for 40 000 keywords. The algorithm then assigns each researcher and keyword a vector, as described on the site : "[E]very author’s vector is close to a weighted average of the vectors of all keywords they use, and every keyword’s vector is close to a weighted average of the vectors of all authors who have used that keyword.” Scimeter then compares those vectors to compose a word cloud of an author’s commonly used terms, to match up similar authors, or to rank authors by their expertise in a given topic.
Scimeter generates this word cloud for the late theoretical physicist Stephen Hawking. Larger words are more frequently used in Hawking’s arXiv papers; red terms were used more recently than those in blue.
Scimeter
Ludo Waltman, deputy director of the Centre for Science and Technology Studies at Leiden University in the Netherlands, says he plans to use Scimeter to find peer reviewers for papers submitted to the journal he edits, the Journal of Informetrics. But he says the functionality of Scimeter for other purposes, particularly identifying similar researchers, is limited. When he specifies two fields of study, the results display researchers prominent in either of the fields but not both. “Scimeter currently doesn’t seem very suitable to identify researchers who establish interdisciplinary connections between fields,” he says.
Information scientist Cassidy Sugimoto of Indiana University Bloomington says that the site tries to fix a problem that doesn’t exist. “Finding scholars who use similar keywords or topics is possible from web searches and major bibliometric databases,” she says. Hossenfelder agrees that commercial publishers and services such as the Astrophysics Data System offer similar tools, but she says the ability to visualize the topics of authors on arXiv is novel.
Early users are also wary of a new metric that Scimeter includes in its analysis of researchers: normalized broadness, a measure of the scope of a scientist’s interests. Scimeter calculates a given researcher’s broadness by analyzing the distribution of that person’s most commonly used keywords. “We assume that broader authors will have a less predictable topic distribution,” Hossenfelder and Scimeter cofounder Tom Price write in an arXiv preprint published in May. Researchers working in areas that have fewer applications in other fields, such as quantum gravity, are likely to have broadness values below zero, the average of all arXiv authors.
Hossenfelder says the broadness metric can supplement simple measures of research productivity like the h-index. Job recruiters, for example, might find the metric useful if they need a candidate with a specific area of expertise or with more interdisciplinary interests. Sugimoto, however, is not convinced by the calculation underlying the broadness metric. “It fails to take into account decades of research on specialization and topic analysis that could provide more robust indicators of breadth and similarity,” she says.
What the future holds for Scimeter depends partly on how much more funding Hossenfelder can attract. The aim, she says, is to make it more user-friendly and add more functionalities, such as allowing users to rank authors on the basis of different metrics and expanding to cover repositories other than arXiv.