* Print the paper out and make notes on it by hand (this has a small mention in one of the articles) - this technique got me through university, I tried every software package imaginable but in the end having a hard copy forced me to deal with sheer amount of reading I had to do. It also allowed me to freely make notes, add sticky notes etc. Since then I have done this with any collection of papers I want to get my head around.

* Try to comprehend something on every read through - it doesn't have to be big, just something - whether it is the sample makeup or part of the methodology or the conclusions etc. You don't need to understand these in any particular order, but ensure to revise your understanding as you become more familiar with the paper.

This is very important advice. Do not assume that just because it's a published research paper it is valuable, correct, or useful. In fact, especially in applied CS, I found that authors will sometimes make what looks like an intentional effort to obfuscate the methods so that the paper is publishable (peer reviewers will not try to reproduce the results anyway), but the methods are not implementable, or at least not easily.

Lately I've been reading "How to Read a Book" by Adler and van Doren [1], which systematises this. It applies to papers just as well as books. Get the 1972 edition, as the 1940 edition is quite a bit less readable.

I would say that most papers are worthless. The publish or perish mentality in academia leads to more noise and less signal. You really have to fail fast on papers, find the seminal ones and the few under looked gems related to your topic that can hopefully be fished out via Google (though word of mouth is important).

The answer is to do it. Do it often, do it a lot, over and over. The linked tweet provided great advice, as do other resources and comments in this thread. But the bottom line is that you learn by doing, and trying hard.

I remember reading the same section of a paper at least 30 times, no exaggeration. I would read it, then go and look up the topic online, then read it again, then go do something else, and then read it again. Over and over for days. i parsed every symbol and word of every line. And then one day I was elucidated. I finally understood it, and it was a feeling of awesome communion with the gods.

I'm not sure if there's a short cut, but I'll be monitoring this thread along with you in the hopes of finding one. Do what the others said and print the paper out, and take a pencil and highlighter to it. Look up terms you don't know online and be diligent in your use of search engines. Just stay at it, and don't give up!

If it's only a dilettante's interest, then read it like you would read news: skim for a general sense, emphasizing the introduction, conclusion, and graphics.

For academics reading papers, the situation is (potentially) different. There are genuine professional consequences for failing to read or understand the material.

There are many approaches -- many good ideas here -- most of which are rehashing of basic study skills, but I would add one piece of generic advice.

Journals have strict page limits so there's no space to introduce domain specific terminology or give a brief tutorial on the topic. They only have enough room to tersely describe directly relevant background material and will assume you are knowledgable in the domain.

I read the abstract, if it seems relevant, then I flip to the conclusion, if it still seems relevant then I skim read it through, then print it out if it looks like something I need to dig deeper into.

First read title and abstract. This is where you'll likely eliminate a lot of stuff.

Interestingly, I've found that the best point to go from there is the conclusion. A good conclusion will generally tell you the results (which is generally what interests you), plus a good summary of the motivation and approach/methods.

I hope that soon reading papers will be supplanted by automated aggregation and parsing of content, and eventually almost real time dissemination of data and results. Frankly, encasing scientific knowledge in papers is now beyond a joke. Why we are still relying on parchment based technology and pre-enlightenment publishing norms drives me nuts. Examples:

1. Genomics papers that have a table with thousands of rows printed in a PDF in the appendix. Almost willfully annoying.

2. Two page papers in top life sciences journals that have 100 page appendices which are not searchable or indexed.

3. Papers that are heavily criticised in letters, but none of this is obvious when looking at the original paper HTML page, the abstract or the PDF.

4. Long clinical trials that are published repeatedly over 20 years, where the previous interim data updates are locked away in journal archives that requires extra payment or a subscription to JSTOR.

5. Papers with 10 supplementary tables, all single file excel spreadsheets.

6. Papers are submitted for publication 12 months before they are actually published. Bad luck if you started working on the same thing 6 months ago.

7. If I want to show a table from your paper in a presentation, I have to manually snapshot from the PDF and paste it on my slide.

8. Because the only way to disseminate data is with a paper, important trial results can be given as a presentation at a conference, and not become available until published much later. Until then, the only record of these results are your memory if you were there, a pay-walled recording of the presentation, a drug company press release, or sandwiched between banner ads on a 'medical news' site.

9. Old papers get ignored, because they just aren't readily available. Life science researchers regularly 'discover' something that was published in the 1970s.

10. It is impossible to read all these papers. One part of my research field easily has 10000 papers or more that could be justifiably relevant. Biology has now reached the absurd tipping point where it is easier and cheaper in terms of labour costs to do an experiment yourself rather than thoroughly survey the literature.

Hopefully computer science papers aren't like this.

Where you focus on the paper depends on what you want: if you are just learning the central questions and theoretical views in the field then the beginning and end are more useful; as you gain an understanding of the field you may be more interested in the particular advances in a given paper and may devote more attention to the middle.

At least in several of the fields I read regularly (linguistics, cognitive science, NLP, and developmental psychology) there are three key paragraphs that tend to have the most useful information: 1) the abstract (duh!) 2) the last paragraph of the introduction (this usually has the most concise statement of the motivation) 3) the first paragraph of the discussion (this usually recaps whatever advances they make in a usable way; actual results sections often contain a level of detail that isn't very memorable after a few days).

Citations are shorthand for entire sets of ideas--they're a means by which whole arguments can be referred to very concisely. One of the challenges in reading a new literature is in learning what set of citations are commonly reused and how, especially because the conventionalized citation of a paper isn't necessarily what the author expected it to be known for, nor what you might take from a paper on a first reading (or any number of readings). For this reason I highly recommend reading papers with inline full citations (Chomsky, 1965) rather than numeric/ indexical citations that you have to look up [1].

If you're reading for expanding your knowledge in a field, then a good area to start is by reading well known dissemination in the field in past decade or so. So, say, you're into datastrucures, read the paper that brought Red-black tress by Rudolf Bayer. This is also a great route for somebody reading for the first time. Well written papers are less off-putting. Most papers out there are very poorly written.

If you're trying to extend the field, then it helps scouting what interests you in extremely recent, interesting works. You can use sites like arxiv to find edgy works done in past month of so, and then drill down the references to understand where they came from. You can then work on improving that work.

So it depends on what you're trying to do. For software, I usually use GoodReader on iPad. Nothing fancy, but it annotates, stores all my research papers and organizes them. On computer, give Zotero a try.

If I decide that i absolutely need to understand a particular paper, I start by finding a relevant review article. I look and see if the papers cited in the background section are also cited in a recent review of the state of the field. These are less likely to be flat out wrong, and they provide better context.

I try to read a paper with other people (preferably people who know the field) and we take turns trying to explain the figures and methods to each other.

But to reiterate: any given paper is most likely wrong and your default position should be persistent skepticism.

2) Be cognizant of absorbtion / attention: Its easy to get stuck glossy-eyed and pacing through the lines

Throughout the years I've collected thousands of papers, many still unread, but my 'must read' ('must skim' mostly) folder currently houses several hundred of them and I can't imagine how much time I would've wasted by manually renaming using the template {Author} - {Year} - {Title}.

Mendeley is free, but a paid alternative is Papers (http://www.papersapp.com/), which used to be Mac-only and the reason I ended up discovering and using Mendeley. Haven't personally tried it, but heard very good things about it.

http://violentmetaphors.com/2013/09/08/an-example-of-how-to-...

And

I personally have to sit down with pen and paper and toy with key equations, maybe take out a book or two to make sure the author and I are agreeing on specific vocabulary, or even double check symbols used which can differ from what one is used to.

And keep in mind that completely novel, truly groundbreaking papers in a field are rare. Most of what I've read is iterations on very well known basic methods, a modification of this or that, some change to XYZ etc.

If you're doing post-grad (or research) computer science, you should really think about learning to read the literature faster. I was very slow at this when I started, and I notice new grad students struggling at it.

After doing that I think about all the questions I have. Then go forth and read it through keying in on the areas that may answer my questions.

I also see a lot of useful tips here such as making notes, rewriting parts of the paper, and so forth. I think these ideas only work when you're already getting something out of the paper. These tips are optimizers, they can help you get more out of a paper if you're already somewhat conversant in the field.

In theory, if you're not an expert, you can keep following the citation trail, read all the papers there, become an expert and finally read the paper you actually want to read. I don't think this actually works. If you're not a expert, the paper itself will seem like gibberish. You won't know what citations are actually important and what citations were just pushed in to please a reviewer. And even if you somehow do, the citations themselves will be just as difficult to read.

So what are the solutions? First, seek out accessible papers. There are a lot of tutorials that get published in magazines like the CACM and IEEE Spectrum. These are written with an explicit goal of being accessible to a technical non-expert audience, so you will get something out of these articles. Maybe if you read one such article, you'll get the big ideas in the article and then you can read the technical paper(s) that the article was based on. Since you already know the big ideas, the technical paper won't seem so daunting. And maybe now, you can follow up on a few citations.

The general trick is to find alternate sources that simplify and explain the paper and then read the paper. Alternate sources can be tutorial articles, slides, blog posts, textbooks, talks and so on. This is actually how people learn to read papers in grad school. First, you go to few a masters level classes where the professors teach from papers rather than textbooks but don't necessarily expect students to read and understand all the details in the paper. Next you go to a few seminar-type classes where professors and students discuss the nuances of a number of important papers. There's a lot of non-obvious spoonfeeding that happens: the professors use their expertise to pick papers that important and accessible, they guide the discussion towards the important parts of the papers, they point out flaws and discuss methodologies, suggest projects for the students to implement and learn from, and so on. Over time, the students absorb these lessons by osmosis and become good at picking out and reading papers themselves.

* Take a sheet of paper and write down all definitions for easy reference

I believe the key statement that has been iterated over and over here is that only 10% of scientific papers are ever worth reading. And probably only 10% of the text of those 10% are worth reading. They are they 1%.

This means that you end up with hundreds (if not thousands) of papers that you think you might have to read, but have no way of knowing whether or not you should until you have read them. Chicken meet egg.

I built Qiqqa (and wrote a PhD) to solve this from two broad directions: using human intelligence and using artificial intelligence.

USING HUMAN INTELLIGENCE: While a load of people recommend printing out a paper to read it, I believe that that course leads to a lot of future pain. Highlighting and annotating while you read is absolutely important. It allows you to skim through a paper and highlight only the important stuff. Then you can come back later to properly read the important papers – only once you have skimmed enough papers to get a more general sense of the domain you are exploring. If you have printed out papers, it is very difficult to quickly regroup and reread your annotated papers. However, if you have highlighted your papers on your computer, laptop, tablet (or even phone while travelling from Cambridge to London), free products like Qiqqa (or www.pdfhighlights.com) offers you a simple annotation report to pull out all your annotations to not only remind you of what and where the important fact are, but also to let to jump straight to them to get reading immediately.

USING ARTIFICIAL INTELLIGENCE: An interesting side effect of reading more scientific research is that the more you read, the more you have left to read. Every paper you end up reading will make reference to another few papers that you probably should read; like The Magic Porridge Pot. To solve the problem of ‘what should I read next’ (the working title of my PhD), I built into Qiqqa the capability for it to ‘automatically read the papers for you’ using machine learning. To this end, when reading a paper, Qiqqa can recommend to you the most relevant papers to read using something called Topical PageRank which calculates the relevance of a paper to what you are reading not only by its similarity to what you are reading, but also by how well received (cited) that paper is in your community. Think of it as having your own personal Google where it biases results to satisfy your personal predilections. You can read about how it works at http://aclanthology.info/papers/topical-pagerank-a-model-of-....