Libraries and Learning
Inevitably Open
August 1, 2017
According to a new study uploaded in preprint form to PeerJ and profiled in Science, subscription-based publishing is a doomed business. Scientists have flocked to Sci-Hub because it’s easy to use – no traipsing from one site to another, no fumbling for a credit card or a delay while an interlibrary-loan request is processed. Just drop in a DOI and there’s a good chance you’ll get the paper instantly or, if it’s not already in Sci-Hub, very shortly. The chances papers will be there are greatest for the largest publishers, including those that have brought suit against Sci-Hub. (Interestingly, the researchers charted Google searches for the site’s name and saw they spike every time a suit is brought against the pirate aggregator.) The conclusion: “this is the beginning of the end for subscription scholarly publishing.”
Of course, there is one problem: libraries have to subscribe to these journals in the first place for them to end up in Sci-Hub, which relies on borrowed (or phished) library login credentials. What the popularity of Sci-Hub has demonstrated, though, is that the limited access libraries have to offer is less effective and efficient than what can happen in a less copyright-hobbled world. How are you going to keep the scientists down on the intellectual property farm once they’ve seen how much easier it could be?
Another recent Science article (hat tip to my friend Larry for pointing it out) reports a study conducted by Danish researchers (posted to the open repository BioRxiv) that found you can learn a lot more by data-mining the full text of a large numbers of scientific papers than by simply analyzing a lot of abstracts, which is what you do when you can’t get hold of the actual papers. The researchers created two pools of data: one, abstracts of over 15 million articles, the other the actual texts of those articles. I’m tired just thinking about it. They faced two big challenges: getting the publishers’ permission to mine their articles and making PDFs consistently machine-readable. They’re related issues: journal publishers use different formats which makes conversion to text files labor-intensive, and (no surprise) they aren’t willing to let researchers share the articles that have been painstakingly readied for text-mining. Though 15 million is a lot of articles, the authors drew on only three sources: the backfiles of two large publishers (Springer and Elsevier) and PMC from the National Institutes of Health. It would be incredibly valuable to be able to text-mine vast amounts of scientific knowledge, but that’s tricky when so many publishers are involved using different formats who have little incentive to make the literature available for mining. It would be sort of like mapping the human genome, but with more intellectual property lawyers.
I have no idea where we’re going with this. If the SciHub model succeeds, it will kill its host. Open platforms like ArXiv, Humanit