I'm working on a deep neural network research application Does usage of youtube videos or google images as data source cause a sharing-data, ethical, copyright or commercial problem? The aim of the usage of these data is only academic research. Maybe we can neglect google-images because of there are many free image data set. But we need youtube videos because of time and cost. Instead of sharing data in the article, is sufficient to write: "this study uses top 10(or more) search results for keyword search on youtube/google". There are lots of machine learning examples that learn from google like https://www.pyimagesearch.com/2017/12/04/how-to-create-a-deep-learning-dataset-using-google-images/
5
-
1One important aspect is that both Google and Youtube search results are personalized and change with time, so others might find it difficult to reproduce your results. This can probably be solved by keeping a copy of the training set, and being prepared to share it, but it's worth considering if another source would be more stable.– AnyonCommented Nov 13, 2018 at 14:41
-
@Anyon, keeping and sharing is precisely what may not be allowed under copyright law. Giving the time of access is enough to establish validity. In fact any cited reference to the internet should similarly be dated (and timed) as websites can change and you don't have control over that. If results depend on one specific data set then they have little true scientific merit in any case.– BuffyCommented Nov 13, 2018 at 18:52
-
See Copyright – use of images to train algorithm– ff524Commented Nov 13, 2018 at 20:30
-
@Anyon : It is a good point. The customization of the data according to the personal search will not create a problem in terms of the study aim because the study will only reveal the theory of such an application. We are not looking for a commercial performance or standard. A few different algorithms will be compared and only prove of a working system.– acsCommented Nov 13, 2018 at 21:07
-
@Buffy Your point about the legal matters is well taken. To what extent the specific data matters will of course depend on the details of the study, and the nature of the reported results. For sufficiently qualitative work it might not matter much at all.– AnyonCommented Nov 13, 2018 at 21:52
Add a comment
|
1 Answer
I think this depends on more than what you state here. In particular, it depends on what you intend to do with the data you analyze and how you intend to present it. If you copy/replicate/publish the material, then you need to deal with copyright as anyone would. But if you are only extracting naturally anonymous information from it you should have no problem. Your suggestion that you "reference" your dataset not by providing a copy of it, but by giving (with the date of access) the search terms that generate it, you should be fine.
As an example, if you are researching how many people pass by a certain street corner on a given Tuesday, you can gather the counts without getting "informed consent" from everyone that passes. No issues there. But if you intend to publish, somehow, the names (or images, or other information) of those people then it is a completely different story. You will need to judge where on that spectrum your study lies. I think you can even categorize things in the aggregate as long as you don't label individual items with your categories. To say that "60% of the videos we examine have blue backgrounds" is benign. Even saying "40% of the videos in our sample present conservative positions" is fine, as long as you don't "name names" and point fingers. (I am not a lawyer, of course.)
If you were in the US, then your IRB would give you guidance on this. In your country there may be institutional rules to guide you. You should seek those out. But generally, the "academic use" exceptions to copyright that were in place in the past have been fading out as copyright holders try, usually successfully, to extend their control over the culture. Not a good thing, but the law can be used against you.
