Being a journal reviewer is an important part of scientific activity.
In spite of Journals' suggestions for the reviewer, the process seems to
me a little exhausting, especially in computer science.
The reason for me is how to ensure:
- reproducibility of the exposed experiment
- data trustfulness
- validity of the method.
Of course the way to fully prove the idea of the paper is to re-implement the experiment with the same and other data. However, many times some reviewers leave to their experience what seems to be "plausible" to be "correct". But in my case I am even tempted to program some little test. Does anyone have experience to share on this?
In reviewing process rigor and efficiency should be combined. It is the reviewr level of expertise which will make the difference.
I think that what you're looking for is reproducible research. Claerbout, then Buckheit&Donoho and others make the point that an article is an advertisement of scholarship, not the scholarship itself. By providing the program in a (hopefully) free implementation of whatever programming language is used, the scholarship in itself can be examined. Unfortunately it hasn't spread so much as it could and should.
As reviewer, you can also suggest to authors to apply online demonstrations of their code.
A good way for such purpose is the IPOL journal (www.ipol.im) which now purposes to construct on line demonstration (see. http://dev.ipol.im/ws/) to convince reviewer and demonstrate the reproducibility of experiments. Don't hesitate to propose them to author and it is free they just have to give their code and compile on server.
Good luck for your review and hope that it will help you.
- Badges
- Science topic
- Similar topics
- Computer Science