Anatomy of a Scientific Research Paper (Part 1)

I was in school for 23 years.  And the hardest thing I ever learned to do was write a measly 10 page research paper.  I had my research results.  I had built my tools.  I had conducted all the necessary experiments.  And yet the writing seemed impossible.  Sentences came slowly, if ever.  Every word was painful.  I would do anything else, and chores became a blessing: clean my apartment, scrub the dishes, go to the dentist, anything but write.

I thought I might not be cut out for research.  Maybe all this school was not for me after all.  It’s fashionable to quit nowadays.

Nonsense.  What was wrong was that I didn’t know the anatomy of a scientific research paper.  I didn’t even know where to begin.  If this sounds familiar to you, read on.  Over the next few posts I will spell it out for you one \section{} at a time.

Let’s start with a bird’s eye view of a typical 10 page technical paper.

1-2    1-2
Pages 1 and 2.  This is where you describe the problem you are solving.  This is not where you describe your solution to the problem.  You may hint at the solution in the last paragraph or so of the introduction.  But a vast majority of the space on the first two pages should be dedicated to the problem.  The introduction should give a general description of the problem domain.  The “The Problem” section should give background, such as the source of the problem, the negative consequences of the problem, and the potential benefits to solving the problem.  In other words: why is the problem important?

3-4    3-4
Next up are pages 3 and 4.  Page 3 is the background.  This means any tools and technologies that you rely on for your solution.  This is not where you describe your solution.  This is only where you give an overview of the most-important technologies you use that have been published before.  If you are proposing a new feature location technique that combines Latent Semantic Indexing with Program Slicing, you should give a brief description of these in the background section.  The background is not for rehashing the problem.  And I repeat, the background is not for the solution.  It is for the supporting technologies of the solution.

Finally on page 4 is where you start explaining your brand new, enlightened solution.  Provide a picture to show the architecture.  Make it 100% clear how you combined LSI and slicing.  What, precisely, are the inputs and outputs of your solution?  What parameters did you use?  What design decisions did you make and what was the rationale behind each decision?  Etc.

5-6    5-6
Now we come to pages 5 and 6.  Page 5 is a continuation of the approach section.  This is also a great place for a concrete example.  Pick out one example and explain it thoroughly.  Show exactly what the input and output of your approach.  This example will lead nicely into the evaluation on page 6.  Now, here you do not include anything about your research results.  You only talk about your evaluation methodology.  I repeat: you only talk about your evaluation methodology, NOT THE RESULTS YET.  By methodology I mean, what gold sets did you obtain?  Did you have human evaluators?  If so, how many and who were they?  What metrics did you use?  And above all: what were your Research Questions?

7-8    7-8
Page 7 continues the evaluation section.  Remember to end the evaluation section with a subsection called Threats to Validity, to let everyone know what you perceive as the weaknesses of your study, and what you tried to do about those weaknesses.  On page 8, at last, we get to the Empirical Results.  These are the results of your evaluation that can be quantified, such as metrics you measured or time taken.  This is also where statistical test results go, and is usually where you answer most of your research questions.  This is not where you describe the problem or the solution, or your evaluation methodology.  It is only where you describe your quantifiable results.

9-10    9-10
Page 9 might be some spillover results, and also any qualitative results you have.  Qualitative results being subjective results such as comments from human evaluators or unbiased observer opinions.  Page 9 and 10 will usually also have your Related Work section, though some papers include this up front.  And of course page 10 closes with a short conclusion, acknowledgements, and your references.  Sometimes conferences allow the references to fall onto an 11th or 12th page, provided you don’t put any content on those pages.


Whew!  That’s a lot of writing.  But it is also doable if you break it down into these digestible nuggets.  Work a little bit each day, and try to write one section per week.  That will get you a first draft within two months.  If you are smart, you will be writing while doing the research, so you will have results just in time to write the results section.

One more thing to notice here: we only spent about 2 pages on the approach.  We spent two pages on the problem, one and a half on the new idea, and then four pages on the evaluation and results.  Got it?  That’s just 20% of the paper on your new idea.  The other 80% is spent talking about the problem, supporting technologies, related work, and evaluation.

Leave a Reply

Your email address will not be published. Required fields are marked *