The title is link-bait. That really isn't what Elsevier is saying here. They are complaining that this researcher broke the terms for the university's subscript

The title is link-bait. That really isn't what Elsevier is saying here. They are complaining that this researcher broke the terms for the university's subscription. On their website they state: [0] """Further, we're looking at how we ensure that researchers know what they can and cannot do with content, or where to go for further information, without giving the impression that we are claiming ownership over non-copyrightable facts and data.""" Regarding this case, (IANAL) it seems to be that if their APIs worked properly, then Elsevier is completely in the right (especially in the UK) in demanding that papers only be accessed through the API. If they don't work, as the researcher alleges, then would I agree that's he's in the right. A couple weeks ago I was reading all about Elsevier's APIs for downloading papers [1]. If you're at an institution that has access to ScienceDirect or Scopus then it seems you can easily get a key that gives you full access to everything, including papers in XML instead of PDF if you want (eg. a mathml-like for equations, paragraphs, figures, tables all marked up). However Elsevier make it very difficult to find the actual terms and conditions for text mining on their website, despite numerous pages which run you in circles. They are here [4]. To summarise them (again, IANAL): non-commercial research use only. You can't share the raw text mining output except with other people belonging to your institution/subscriber, although you could allow indirect access through e.g. a website. You can also distribute short snippets from the text with a copyright notice. You can keep the downloaded data until your API key expires, which happens if you stop using it or your institution stops subscribing. I also found it strange that Elsevier forbid redistributing the abstracts of papers [3], considering that they are publicly accessible. It would be fantastic if publishers couldn't restrict the use of the information in the articles they published, even for commercial use. Anything that any human learns from a copyrighted work can normally be used without restriction, why wouldn't the same apply to machine learning? However I assume that when you sign an agreement providing access to data (e.g. subscribe to Scopus) that you can sign away that sort of right (e.g. NDAs). The Hague Declaration [2] is a great initiative in this direction but unfortunately it's only petitioning for the rights of researchers. [0] https://www.elsevier.com/connect/how-does-elseviers-text-mining-policy-work-with-new-uk-tdm-law https://www.elsevier.com/connect/how-does-elseviers-text-min... [1] http://dev.elsevier.com/ http://dev.elsevier.com/ [2] https://creativecommons.org/weblog/entry/45456 https://creativecommons.org/weblog/entry/45456 [3] http://dev.elsevier.com/policy.html http://dev.elsevier.com/policy.html [4] https://www.elsevier.com/__data/assets/pdf_file/0012/102234/TDM-sign-up-short-form.pdf https://www.elsevier.com/__data/assets/pdf_file/0012/102234/...