At the end of last quarter, I helped review a research paper submitted to an upcoming database conference. It was interesting to get a perspective on the review process from the point of view of a reviewer. And it was also a challenge, as I had to read a paper on a topic I had not dealt with much before (probabilistic databases) and provide my judgement on the value of the paper. Certain things, like prose clarity and polished presentation, are easy to evaluate regardless of the subject of the paper, but it’s another matter to attempt to measure things like originality of contribution or scope of impact without prior knowledge of the topic. In order to provide a valid opinion on a new paper, it is necessary to read papers already written about the subject.
Driven by this need, I found a very helpful paper aimed explicitly at describing an effective, efficient technique for approaching scientific papers, How to Read a Paper by Professor Keshav at the University of Waterloo. He describes a three pass approach, where you first spend 5-10 minutes quickly scanning the paper, then read it in greater depth but skip the proofs and technical details, and finally a third pass where you go through the entire paper, carefully reconstructing each argument yourself. He also provides a very helpful description of how to begin a literature review, wherein you compile a summary of a collection of important papers on a particular topic.
The three-pass approach to reading papers has been very helpful to me. Previously, I approached every paper with the same method I take when reading a textbook: read with pen and paper in hand, recreating each proof and main argument as you go. For me, this is the most effective way to obtain a full understanding of a paper’s contents, but it takes a significant amount of time. It is also a very inefficient way to approach a paper that you may not even be sure you need to read in its entirety. The sequential passes, in increasing level of depth, recommended by Professor Keshav let you determine exactly how detailed an understanding of the paper you require.
In addition to this advice on reading papers, Professor Keshav maintains an interesting collection of links to advice, both written by himself and others, covering all manner of subjects related to computer science research, including but not limited to the section “What to do when you don’t know anything”.