1

I am working on a paper after peer review and struggling to address one of the comments that states "empirical method without any solid theoretical background". It is a paper about machine learning algorithms applied to a specific field.

Please, any ideas or suggestions on how to address this comment?

I reviewed my methodology section and brought more formalism to the design decisions. Additionally, I am working to address all the aspects of reproducibility as the answer to question Guidelines for communication of machine learning methods in empirical research papers.

5
  • Is it your understanding that you are using standard ML procedures?
    – Buffy
    Nov 22 at 15:21
  • 3
    I think it's rather hard for random people on the internet to contribute positively by speculating on appropriate revisions for a paper they'll never see. Nov 22 at 15:32
  • @AnonymousM I'm struggling to frame my understanding on how to give a solid theoretical background for an empirical method.
    – Eckz
    Nov 22 at 21:23
  • @Buffy, yes I'm using common practice on this domain, also based on my related works.
    – Eckz
    Nov 22 at 21:23
  • 1
    Then my best guess is that the accepted answer is correct. The reviewer knows the domain but not ML processes (and maybe is a skeptic). Talk to the editor.
    – Buffy
    Nov 22 at 21:28

1 Answer 1

3

We have nowhere near enough information to tell you anything for certain. But given what you write, I will make a couple of assumptions:

  1. With "specific field" you mean some substantive subject like migration or labor market
  2. You are a specialist in Machine Learning, but not a specialist in the specific field
  3. The reviewer is a specialist in that field, but not in ML

If that is the case then the reviewers comment probably means that (s)he misses a description of the theories for that field (e.g. why do some people migrate and others not), and a discussion on what your results/methods can tell about these theories. In that case more formalism and improving your methodology section is not going to help. In that case you will probably have to bring in a subject matter specialist as a co-author.

Again, this is based on very little information and a lot of assumptions.

4
  • 1
    This was also my first guess about the situation. However, the editor might be able to resolve it without bringing in another author. It would depend on the journal and the expertise of the editor.
    – Buffy
    Nov 22 at 16:37
  • Thank you very much @Maarten Buis. I am tending more to hypothesis 3 (based on other reviewer' comments). I will do my best and let the editor decide as pointed out. Additionally, I found a nice publication that discuss about the experimental works, and also this tweet... It is an ongoing discussion for works with ML.
    – Eckz
    Nov 22 at 21:31
  • 1
    My answer rests on all three assumptions being true, not just one of them. Nov 23 at 11:09
  • @MaartenBuis thanks. I had not considered hypothesis #3 previously, and this gave me insight into Buffy's comment. Regarding #1 and #2 it is not the case (and there is a background section to address both).
    – Eckz
    Nov 24 at 11:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .