Research Papers Shouldn’t Be Read in Order; or, How to Read a Research Paper

I just finished a blog post where I discuss things I’ve recently learned about how to read research papers. I almost included this as a point in that post, but I think it’s important enough to warrant its own article.

Here’s the idea: you absolutely should not be reading the sections of a research paper in order.

It took me a while to learn this one – I can’t remember if I first read this advice somewhere, if someone told it to me, or if I reverse-engineered it from advice I got about how to write papers. It doesn’t matter which one came first, really – my point is that I only found this out in a roundabout way through googling and trial and error and harassing people with questions.

(Seriously, why isn’t teaching students how to read papers a thing we do in undergrad??)

I should probably point out that my experience is based on reading math and theoretical computer science papers. I have no idea what is done in other fields, so your mileage may vary.

The correct order in which to read a paper

I typically read books – even nonfiction books and assigned reference texts – from cover to cover. However, reading a paper from cover to cover is unideal. This is the order of sections in which you’re supposed to read a paper:

Title
Abstract
Introduction
Related Works/Background/Preliminaries
Conclusion
References
Body: Methodology, Results, Discussion, etc. (skim)
Body (in detail)

This is the ideal order in which to read a paper because it takes you from the most general view of the new idea to the most technical and in the weeds view of the new idea. Depending on your reason to look at the paper, you might never need to get all the way down to step 8 (detailed reading of the body).

It’s also good to note that it’s most effective to take multiple passes to fully read a paper, and so at each pass, you might want to stop at a different level of depth. This order provides a lot of natural points at which to stop reading a paper.

The title should tell you something about the topic of the paper, so if it’s completely irrelevant you might be able to skip reading it (this mostly applies to trying to figure out which referenced papers to read next). If you’re just browsing through papers, you can probably stop after anywhere between step 2 (reading the abstract) and step 6 (skimming the references).

Incidentally, you’re supposed to write a paper in roughly the opposite order.

The Abstract

The abstract is a condensed summary of the entire paper. It’s worth making note of any new ideas or key terms in the abstract and coming back to them after skimming the introduction and background, or even setting it aside to learn some prerequisite material and only revisiting the paper once you understand the abstract. If there’s any part of the paper you should aim to fully understand, it’s the abstract. If you can understand a paper’s abstract, you’ve gotten the tl;dr of the idea and the results.

By the way, understanding abstracts can be involved and difficult. I recently did a presentation where the main goal was to get my classmates to understand the contents of an abstract. I spent 20 minutes on background and contextualization, and barely got any further than material from the abstract. This is not a joke.

The Introduction

The introduction section is usually an extended version of the abstract, where all of the ideas are presented, but perhaps without all of the supporting details. Typically, the introduction will try to give some reasons as to why the research problem is relevant or important, give an overview of some related important ideas, present a condensed version of the new results and ideas with brief references to the methodology or prior results relied upon, and summarize the research questions investigated, as well as the contributions of the paper.

If you get through the introduction, you should have a pretty good idea of what the authors are claiming. You might also have a bit of an idea of why they think their ideas work, as well as some clear jumping off points from which to go looking for more background information (if you need it)The title should tell you something about the topic, so if it’s completely irrelevant you might be able to skip reading it (this mostly applies to trying to figure out which referenced papers to read next)..

Personally, I find introductions to be much more difficult to read than abstracts, because they try to cover a lot of information in a very short amount of space, they skip details, and they assume a high amount of background knowledge from the reader. Also, if only 20% of the paper is going to be relevant to me, that probably maps to only about 20% of the introduction being relevant, and I usually have no idea which 20% that is.

That being said, I can see why an expert or someone with the right background knowledge might find it sufficient to stop after reading the introduction. In any case, reading the entire introduction is good for putting the entire paper into context.

I think it’s good to read the introduction a few times, at least until it doesn’t feel completely foreign to you. However, I don’t think fully understanding the introduction is that important. Most of it will likely end up being background information you don’t directly need. The main things you need to know from the introduction are the research questions and the contributions.

Having an introductory section that kicks off the paper is pretty standard, but papers differ quite a bit in what the next section is. Sometimes it’s a related works section, but it’s just as likely in my experience for related works to be discussed in the introduction. Sometimes, you get a background section, which might be a cross between a “motivation” section, a “related works” section, and a “preliminaries” section. Sometimes motivation and related works are separate from preliminaries or preliminaries are omitted. Sometimes you get a “preliminaries” section all on its own. It really depends on the authors.

In my experience, it’s good to skim the related works section in case there’s anything interesting, but not spend too much time on it. If there is a lineage of other papers that the paper you’re reading is clearly building on, you might want to take note of them to refer to later. Otherwise, the related works section is where authors signal that they’re experts who should be taken seriously by making it clear that they’re aware of prior work done in the field. That prior work may or may not be relevant to your purposes.

Preliminaries sections are gold, if you get them, and worth reading extra carefully. This is where new terms are actually explained (wild, I know). They usually read more like review texts intended for people who already know the material, so if you’re completely lost, the preliminaries might not be enough. But any concept in the preliminaries is an idea that the authors deemed key to understanding the paper, so make sure you’re solid on those!

The Conclusion

I find that conclusion sections of papers are often quite short, and don’t have a ton of analysis or new information in them. The conclusion will likely restate the contributions and the answers to the research questions (i.e. the results) in a few sentences, with maybe a bit of extra information than the introduction.

However, what makes conclusions different from other sections in academic papers is that they are often more speculative in nature. The authors might comment on where they think their results/technique could be applied, suggest directions for further research, or make predictions or conjectures based on their work. It’s fitting: introductions start by looking to the past, and conclusions end by looking to the future.

Reading the abstract, introduction, background, and conclusion is probably enough for a first or second pass through the paper.

The References

If you’ve decided you want to dive deeper into a paper, looking at the references could be a good idea. If you determined earlier that the paper you’re reading is directly built upon some other papers, now might be a good time to find and skim through them. You might skim through some related papers to get a sense of what the commonalities in terms of methods and terminology are. If there are authors whose work seems to be cited quite a bit, you might look into their other work to see if it’s relevant.

Really, this is a sidequest. But I like sidequests! They’re fun and (sometimes) useful.

The Body

You might never actually need to understand the body of the paper in depth; sometimes skimming it is enough to get the information you need. But if you do actually need to understand the body, good luck! This is the hard part. (It’s also where my advice runs out, because I am not very good at this yet.)

Every time I’ve tried reading through the body of a paper, I’ve ended up going through it in multiple passes. This is where breaking things down can be helpful; where splitting up the paper into chunks to read at a time can be helpful; where active reading techniques and coming up with examples can be helpful.

The point is to take the knowledge that’s in the paper and make it your own. This could mean trying to write your own proofs of claims, or filling in details in provided proofs; it could mean implementing some described algorithm; it could mean replicating some method or experiment. This takes forever, yes! But I think it’s the best way to make ideas sink in. Think of it as trying to learn from an extremely terse textbook that makes you derive assignments and readings by yourself.

Except, you know, that you might end up questioning the validity of that textbook, but hopefully that doesn’t happen.

Some other thoughts

I’ve also been told that when reading a paper, it’s important to take some time to critically evaluate it and reflect on its strengths and weaknesses. Maybe future me will have more to say on that – current me is struggling to read and understand academic papers in the first place, let alone evaluate and have opinions about their quality. I’m just a confused undergrad, guys. (Also, I suspect most of the papers I’ve looked at thus far have actually been high-quality papers.)

I’m aware that if you look elsewhere on the internet, you might find some similar advice. But I find that researchers often don’t explain how they do things, let alone why they do things a certain way. I’m really striving to fill in that gap – giving my perspective along with some concrete examples – so that hopefully other people don’t have to go through the same kind of confusion that I went through.

Also, writing my thoughts like this is good for information retention. So there’s that, too.