I read a huge number of papers (sometimes more than 100 a day, getting them through searches in scholar.google.com) and need to quickly get an overview to find out which ones are worth reading in more detail.
First of all, most papers don't need to be read in detail (at least not immediately), as there is lots of repetition in the collection of all papers on a subject. It pays to find out which papers contain the wanted information in the to me most accessible form.
Therefore, I generally read the introduction, which gives an overview of the work and relates it to other (often more readable) sources (which I then try to find if they look promising from what is said about them and/or their titles if given), and introduces important catch words (that can be searched separately, or in Wikipedia to find their precise meaning). Then I leaf through the paper (concentrating on the initial paragraphs of each section, and on figures and other conspicuous things) to find out how easy it would be to read, and what sort of background is needed (more catch words) to understand the content. 80-90 percent of the papers don't survive this stage.
This takes 5-10 minutes per paper - except for the more readable ones, which take longer since I continue reading as long as I understand at least half of what is there and are not bored. I print and annotate the more interesting papers to be later able to find quickly what I want to remember. For papers that look good, I also try to look at the papers citing them (from google scholar). If the topic is new to me I may need to hunt in this way through a web of 20 or 30 papers, before I find the one(s) really worth reading, being clear and detailed enough.
At times I discover that I need to polish up some background reading necessary to understand the topic of interest, and then hunt for corresponding papers. Figuring out what prior work is needed to make a particular paper readable (because it uses the notation, concepts, terminology, or properties) is much easier than understanding the paper itself, and it saves a lot of work, as approaching a topic without having understood the tools is slow and unrewarding. I learn to know what I do not yet know well enough, and I practice that first.
Usually, more papers worth reading survive than I can read throughly. I take the one from the heap that looks at the moment most attractive, and try to read the section(s) containing the core of what I want to understand. The main work is establishing a dictionary between notation, terminology, and concept definitions in the paper and my own way of understanding things (which is sometimes very close, sometimes very different). Then I fill in new information from the paper, rewritten in my own style. This rewriting is what makes the material my own and hence ''understood'', connecting it to the body of well-understood theory that makes up my knowledge.
But understanding comes at different levels. This means that I read many papers several times in different depth, separated by days or weeks or months where I work on different aspects of the problem or on altogether different things. In this way, really good papers are nourisdhing for a long time. I aim for complete understanding only in topics where I believe that complete understanding is around the corner with a limited amount of effort. Nevertheless, partial understanding accumulates and sooner or later becomes complete understanding, as long as enough attention is spent on a given collection of material.
Q&A (4871)
