David Newman, Google
David Newman explained that interdisciplinarity, or collaboration across disciplines, has been measured in a variety of ways. For example, researchers have tracked citations and mapped them to identify the diversity and coherence of a particular portfolio of work. This kind of work has been helpful in demonstrating how interdisciplinary collaboration is rewarded, or not, in academia. It can also be used to trace the extent to which novel disciplinary collaborations can be found.
Another approach is topic modeling as discussed by Nichols above. Its advantage, in Newman’s view, is that its algorithm uses data drawn from the research itself, rather than from institutional structures through which the research was produced, to produce categories and a picture of a body of work. “Topic modeling learns from bottom up,” he noted. “It learns using the discourse of the investigators themselves.”
The tool also can produce an interdisciplinarity score, Newman explained. The topic model identifies the top four topics for particular research and can assess how semantically different those topics are and how novel the interdisciplinary collaboration was, expressed numerically. These scores can be plotted to show the patterns for a particular body of work. This analysis revealed patterns in the work funded by NSF. For example, such programs as Antarctic Earth Sciences, the Continental Dynamics Program, and Sedimentary Geology and Paleontology ranked very high for interdisciplinarity, while Human Cognition and Perception, and General Age Related Disabilities Engineering ranked on the low end. The NSF has programs designed specifically to support interdisciplinary work, and the analysis showed that these programs did indeed produce work that scored high for interdisciplinarity.
The topic-modeling tool can be used to assess a body of proposals or grants awarded, by awarding entity, topic, publication etc., Newman noted. It complements citation analysis and helps investigators avoid pre-defined subject categories that may obscure interdisciplinary collaborations.