9

Discussing the related work is expected in many (if not all?) types of research papers. In the related work section, the author explains the relationship of the current work to previous work and provides an overview of the relevant literature to readers who are not so familiar already with the research direction to which the paper belongs. As a secondary purpose, with the discussion of the related work, the author also shows that she/he knows about the field.

There are however soft limits to the length of such a section, stemming from the expectation that a substantial part of the paper's pages are used for the actual contribution. This means that only the most relevant related work can be discussed. And what is most relevant is decided by the author.

Now a common criticism in paper reviews, often combined with a rejection or request for major changes in the paper, is that seemingly relevant related work is missing. Typically, the papers mentioned by the review are related, but not infrequently at a level for which 100s of other similarly related papers can be found.

Are there strategies to defend against such criticism, provided that the related work section is quite polished?

The question is particularly relevant for work that uses concepts or builds on approaches from multiple streams of research, which all need to be discussed.

Just resubmitting with some discussed papers replaced by the reviewers' requests until the paper is accepted probably wastes reviewer and author time. Extending the related work section to cover >20% of the paper seems excessive (at least in CS in papers presenting new approaches to solve some problem) and invites criticism about a low density of novel content. Adding additional related work to a not-published appendix may look odd and invites the criticism that it should be part of the main part of the paper.

6
  • 11
    "Now a common criticism in paper reviews, often combined with a rejection or request for major changes in the paper, is that seemingly relevant related work is missing." That isn't my experience at all. None of our publications so far has been rejected for this reason and I have never recommended rejection for this reason. This probably depends on your (sub)field. Is your field CS? If so, please indicate with a tag and also at the beginning of your question.
    – Roland
    Jul 20, 2021 at 10:39
  • @Roland While I am particularly interested in the CS case, this question is not actually specific to CS. I deliberately did not write that this is about papers primarily rejected for missing citations. I've seen the missing citation argument both on the program committee member side of CS conferences as well as from the author's side to round off lists of points against accepting a paper. More than once, the missing citations were the only seemingly objective point against a paper (others being contribution being too small, problem solved not interesting enough, etc.)
    – DCTLib
    Jul 20, 2021 at 10:55
  • @Roland That probably means you're already implicitly using the sort of strategies that the question is asking about...
    – jakebeal
    Jul 20, 2021 at 13:51
  • @jakebeal I suspect it is just easier to identify the most relevant publications in my field. Also, in more experimental fields, peer review has a stronger focus on methodology than on giving an overview of related work.
    – Roland
    Jul 20, 2021 at 13:59
  • @Roland I think of that less as a property of experimental work than of study of existing systems vs. construction of novel systems.
    – jakebeal
    Jul 20, 2021 at 17:37

1 Answer 1

21

Personally, I approach this question by explicitly framing related work in terms of categories and examples. In this manner, as long as the related work is within one of my categories, it is covered by the description even if the specific work is not cited.

For example, in a paper about an improved algorithm for widget sorting, I might write something like this:

Most prior approaches to widget sorting fall into two categories: weight-based and optical. Weight-based sorting (e.g., [1],[2],[3]) works well for widgets of the same size but different materials. Optical sorting (e.g., [4],[5],[6]) works well for widgets with different sizes but the same material. Neither performs well, however, at distinguishing both characteristics at once. In recent years, several alternatives have been proposed that handle both, including Gruber and McClane's approach based on high-powered lasers [7] and Voldemort's alchemical method [8], but to the best of our knowledge none of these alternative methods can be applied at low cost for high-throughput manufacturing.

It's even better if you can find a recent review paper that provides these categories for you, so that you don't have to defend them, e.g.,:

As reviewed in Jones et al. [9], most prior approaches to widget sorting ...

This sort of structure provides a good defense against "you didn't cite my favorite / my work" complaints from reviewers because you aren't claiming to need to cite everything.

From this point, reviewer requests for citation generally fall into three categories for response:

  • Identification of a missing key work or category that you really did need to discuss: Thank the reviewer and add it!
  • Related work already covered by your categories: If it's reasonable, add another citation to your "e.g." lists; if it's pushing it, explain why it's already covered. If the reviewer asks for four citations in the same category, the "e.g." structure gives you a good reason to select only the ones you want as being representative of their list.
  • Dubiously related material: If your categories are broad and comprehensive, you can explain why you believe the proposed material is either covered by one or out of scope entirely. Invite the reviewer to correct you if you are missing something. This puts it into minor revision space at best and is unlikely to be an issue for rejection.
4
  • Thanks for your answer. It tells me that I should probably restrict my question to the case in which there are no revisions and no reponses, as in scientific conferences (common in computer science). Would you mind if I change my question to its case? The second part of your answer looks odd then, however.
    – DCTLib
    Jul 20, 2021 at 11:31
  • 2
    @DCTLib What I've written here is my strategy for conferences too. The only difference is that you don't have to actually write the response to the reviewers (for some conferences: some CS conferences have now started including a response to reviewers phase!). The actions that I take on the manuscript are generally the same, however, since a reviewer is often representative of many readers.
    – jakebeal
    Jul 20, 2021 at 11:44
  • 1
    The 2nd part of the question makes perfect sense even for conferences: a) some conferences now allow a single rebuttal phase before the final decision -- in case your comments will be read by the people making the decision and b) even if your paper is accepted already, this is good advice. Super-related work you missed? Include and be glad the reviewers pointed it out. Unknown example? Include if relevant/representative/space allows. Tangentially related? Out of scope.
    – penelope
    Jul 20, 2021 at 13:38
  • 1
    Gruber and McClane were coauthors?! Brings a whole new perspective on the events that Christmas.... Jul 21, 2021 at 7:29

You must log in to answer this question.

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