Saturday, December 27, 2008

Reading papers - 5 line summaries!

Posted on Thursday Oct 11, 2007 by ACHARYA, MITHUN

Dr. Xie maintains a very nice bibliography on Mining Software Engineering. We read lot of papers, but with time, tend to forget them. How about having a 5 line summary for each of the paper we read as a part of literature survey? I actually maintain a document which does exactly this and find it very useful. So next time I forget whats in a paper, I go to my document and look for the 5 line summary, and I immediately know what the paper talks about. I dont need to read the paper again. Another useful side-effect of this exercise is when you write related work for any of your papers or thesis. In conferences, when you talk to other researchers, they usually ask - "Have you seen paper X? How is your work different from paper Y?" and its bad not to know some really relevant related work!

Most well written papers, can be read in about 15-30 mins and summarized in about 5 lines. In my field, most papers have a motivating example after introduction. For a well written paper, a reader should get the idea of the whole paper when he completes reading the Example section! So the way I read a paper is - read abstract, look at the conclusion, then read introduction (very fast), and then the example section. This process takes about 15 mins. Then I skim through the framework, implementation, and evaluation details. I spend further time on the paper, only on need basis. Then I summarize the whole paper in about 5 lines! During early years of PhD, it might be beneficial to read the whole paper to learn the art of writing papers... but after getting a hang of writing papers, quick paper reading will be a useful skill!


  1. I think this would be a good practice. I would encourage students to give it a try to see whether it works for you.

    I personally think that a student shall figure out the best/optimal way of reading papers, ... that best fits his/her specific situations. I myself didn't use the particular way suggested by Mithun. But I do keep a bibliography for each research area that I intend to get into and categorize existing projects/papers/links. For example, I recently created webs for testing DB apps and SE for HPC

    I don't know why. I usually can remember the key idea or high level idea for a research paper even after I glimpsed it very quickly; I don't need these 5 lines of summary. That is why I think students shall see which best practice would fit them best.

    But I do keep my WebMon to keep track of latest related papers published by colleagues.

  2. Thanks for your recommandations