7

I'm doing a lot of research in the field of Digital Text Forensics (e.g., authorship attribution, authorship verification or author profiling). In this field there are only very few data sets available, which can be used for own research. One of the few important sources is the evaluation lab on uncovering plagiarism, authorship, and social software misuse (named as PAN).

Over the years, I collected texts from many different internet sources (Blogs, Forums, Amazon, News portals, Project Gutenberg, etc.) by my own, which I aggregated and compiled into various data sets that could be very valuable for the Digital Text Forensics community. Since, as a researcher, I am very interested to share data in order to make experiments reproduceable, I'm facing now a big problem, which I don't know how to handle...

How/Can I share my data sets, without worrying about copyright issues?

It should be also highlighted that the data cannot be anonymized, otherwise it would be totally useless for the research community. Hence, I don't know where to start. Of course there are plattforms such as FigShare, where data sets can be published, but if I would do this I would first don't get any credit and second can run in alot of troubles for publishing texts, which are not mine (even though they are publicly available).

Has someone else faced this problem?

  • Is the number of ways in which the data is used somehow limited? You could offer some form of API access which allows people to use the data without giving full access. – sheß Aug 22 '15 at 12:18
  • What do you mean exactly by API access? In a restricted fashion via user/password for registred users? The data itself can be only used in its plaintext form, in order to provide everyone to possibility to extract their own, desired features from it... – Unhandled exception Aug 22 '15 at 14:41
  • This sounds a little self contradictory. You can not share the data if you don't have the right to share it. – sheß Aug 23 '15 at 21:08
  • 2
    Sharing your corpus might be allowed under copyright law as "fair use." Check the Internet for "fair use". There is a lot of information. Consult an attorney if in doubt. Consider the risks of publishing the entire corpus. If the copyright owner of one of the documents complains, can you remove just that one document? If you don't charge for your corpus, it's unlikely anyone can argue they've been financially damaged by your work. – pndfam05 Aug 24 '15 at 20:35
  • 1
    @Unhandledexception : As it sounds like you might be in academia: if you're affiliated with a university, you might also want to check if there's an IRB (institutional review board) aka. ERB (ethical review board) that has to approve 'research', as they might have rules about data release. (and you could be fired if you don't go through them) – Joe Aug 26 '15 at 0:07
4

Without being a trained copyright expert, the crux of your endevour lies in your process of aggregating your text corpus:

Over the years, I collected texts from many different internet sources (Blogs, Forums, Amazon, News portals, Project Gutenberg, etc.) by my own, which I aggregated and compiled into various data sets that could be very valuable for the Digital Text Forensics community.

Your problem

If you want to share the resulting data set with your name on it (and don't risk any trouble), every source you use must be published under a license which explicitely allows its reuse. That means, to be sure, you would have to read the terms/license statement of every blog you scrape. Forums, if they are older, usually don't enforce a permissive license on user contributions. Project Gutenberg usually has an explicit per-book license, and a general license information page. Their content looks rather fine to use. For commercial entities like Amazon, they probably don't have a license that allows reuse without explicit, written permission.

Practical recommendation

If your process of generating your derived data sets (the ones you want to share) from a set of given inputs is rather automated, you could recreate a reduced version that only relies on sources that were explicitely published under a permissive license.

Then, in the end, you must choose a license for your own publication. If your data set resembles a database, the ODbL might be a good choice. And just like you want to be attributed for your work, you should accompany your dataset with sufficient attribution for the creators of your sources.

Yes, this is all very cumbersome, but that's how copyright law is right now.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.