How to write a Survey Paper

So you want to write a survey paper? How do you begin?

What is (not) a survey paper?

Some of the following is adapted from https://www.researchgate.net/post/How_to_write_survey_or_review_papers_and_What_sections_should_be_mentioned_in_such_papers

To answer this question its best to first ask what a survey paper is not. A survey paper is not simply a core dump of a bunch of papers in a common area.

Instead, a survey is a research paper whose data and results are taken from other papers. This means that you should have a point to make or some new conclusion to draw.

You’ll make this point or draw these conclusions based upon a broad reading of previous works. You need to really know the topic in order to have the audacity to claim that a thorough survey of the field. You’ll need to be completely aware of the main themes, directions, controversies, and results in the field. You may wish to email and interview authors of related works to get their opinion.

Writing a survey paper is much more difficult than writing a research paper

You do not simply list prior results. You need to assimilate and synthesize the results. Sometimes you’ll need to address conflicts in notation or introduce entirely new notation.

And, of course, you need to have a point. The point you make will determine the organization of survey paper. The structure of the main sections of the paper will reflect the structure of field. You might consider the following organization:

  1. Simple to complex scale. Maybe there was some seminal invention that people add more and more complexity onto — this is very very very common in AI and ML.
  2. Comparative Analysis. You compare Two or more different approaches to the same problem.
  3. Pipeline Analysis. Many complex solutions require a pipeline that you’ll describe and categorize and annotate.
  4. Disentanglement. Maybe your field has researchers conflating issues that need to be carefully untangled.
  5. Historical. Tell the story of something if its compelling.

You’ll do a good job if you can communicate a perspective and/or articulate the gaps in the knowledge. This is difficult and should probably not be attempted by young scientists or graduate students.

The bottom line is that you need to have a point to make and, conclusion to draw, or some kind of contribution that is not just a list of abstracts.

An iterative process

The following is taken from https://academia.stackexchange.com/questions/43371/how-to-write-a-survey-paper

The point of a survey paper of the type you are discussion (as distinct from a systematic review), is to provide an organized view of the current state of the field. As such, you should not be attempting to cite every paper, but only the ones that are significant (which will still be an awful lot)

Writing a good survey paper is hard, and there really aren’t any good shortcuts: you do need to become familiar with the content of a very large number of papers, in order to make sure that the view you are presenting is sane.

Step 1: Begin by collecting a large pile of papers to survey.

Start by collecting a handful of papers that you are interested in. See who cites them and what they cite.

Step 2: Things of an organization schema.

Based on your experience and a readings, hypothesize an organization schema for the field. What point are you trying to make with this survey?

Start reading (mostly skimming) and organizing your collection of papers you read using this schema, including noting which ones are most important and which do not fit the schema well.

As you find significant numbers of papers that do not fit the schema well, adjust the schema to better fit what you are actually finding and shift the organization of your collection to match.

Step 3: Find new papers

As you continue to read you’ll find papers that cite and are cited by the papers you’ve read. Add these new papers to the “to be read” collection based on the adjusted schema, then return to Step 1.

Convergence.

When the process converges to a stable schema and an empty to-be-read pile, you will have a well-developed view of the current state of the field and be in a good position to write a survey. Note, however, that this may take a number of months.

Hints and Tricks

Use a bib manager. Zotero, Mendeley, etc

Use a consistent bibtex index structure. I use lastnameYearFirstwordoftitle convention

You should follow this process in your PhD study generally, but it doesn’t mean that you have to write a survey paper. A survey paper needs to have something to say; a point to make; or some contribution in the way we think about a thing.