Let’s read 1700 articles at the same time! NLP and Implementation Science

by Future Historian April 26, 2021

Let’s step into a time machine. The year is 2006.

Swedish electronic group The Knife just released their seminal album Silent Shout. The Winter Olympics in Torino just concluded. And a little article by Eccles and Mittman announced a new journal, Implementation Science. 

PubTrawlr is fundamentally an implementation science website. We are trying to improve the uptake of scientific findings by improving how people engage with large amounts of literature.

But the methods we use are really unfamiliar to a lot of people in the social sciences. So, Dr. Victoria Scott and I decide to help things along by turning an analytic lens on the flagship journal, Implementation Science.  To do this, we pulled every single abstract published since Eccles and Mittman hit the shelf through October 2020.  We then applied some basic topic modeling on the abstracts and plotted these over time.

The result, our just-published article, 5335 days of Implementation Science: using natural language processing to examine publication trends and topics.  It’s an open-access article, so you can read it right at the above link.

Victoria and I thought it was important to put this out for a few reasons.

1. The pace of scientific publications is accelerating, and it’s become extremely difficult to stay on top of things. We are going to be increasingly reliant on sophisticated data curation to make sure the right articles are coming to our attention. Victoria and I wanted to start norming this method so that people are more comfortable with the idea of NLP-guided syntheses. I mean, that’s what PubTrawlr’s all about!
2. Qualitative data is everywhere, but those in social sciences don’t often take advantage of it as much as they should. This is because the analysis of qualitative data can take a lootttttttt of time. For the past year, I’ve been picking on a publication by the National Cancer Institute that purports to be a qualitative guide to implementation science but doesn’t even make a passing reference to NLP.  [You can find it at the bottom of that page]. But, if we were able to use NLP methods more effectively, suddenly a whole body of data becomes available: interview transcripts, news articles, social media posts, all sorts of this that help us better under communities and their context.
3. Equity-related research is becoming more and more important as health disparities really confront us head-on. Implementation science as a field is super health-focused. Is the research adequate addressing equity?  (short answer, not at all.)

So check out our paper at this link.

5335 days of Implementation Science: using natural language processing to examine publication trends and topics

in the coming weeks, we’ll be featuring some of the content cut from the final manuscript due to length, like our work in word vectorization and studying the relationships between core research terms.  Until then, crank up The Knife.