Sunday, August 24, 2008

How to read a research paper

Amit Awekar

My reading strategy is largely influenced by following three papers

1. Keshav, S. 2007. How to read a paper. SIGCOMM Comput. Commun. Rev. 37, 3 (Jul. 2007), 83-84. DOI= http://doi.acm.org.www.lib.ncsu.edu:2048/10.1145/1273445.1273458

2. Amanda, S. How to read a computer science research paper

( http://www.cs.sunysb.edu/~stent/howtoreadacspaper.pdf )

3. Michael M. How to read a research paper

( http://www.eecs.harvard.edu/~michaelm/CS222/ReadPaper.pdf )

Keep following points in mind before you start reading any paper

1. Read with a purpose : You don't read research paper just for the sake of reading. If you do, then better read a novel or newspaper. If you want to get idea about research happening in a totally unfamiliar scholarly field then scientific magazines is a good place to start.

2. Time is the key : Everyone can digest any research paper very well. The only difference will be the time required. My advisor might need 30 minutes while I still need three hours. Once gone, time never comes back. Invest your time wisely.

3. Levels of abstraction : Once you go through the abstract you can have some idea about how relevant this paper is for you. You need not understand every research paper you read to the finest details. Sometimes it is enough to understand the problem addressed and the results (if you are reading any general computer science paper). Sometimes you might want to know the solution methodology (if the paper is somewhat related to your area of interest). And there are some papers that you must understand completely (if you are going to build your research based on them).

4. Jigsaw puzzle : Reading research papers is not all that boring as you might think. Once you understand the problem setting and the results, try to guess the solution that authors have presented. Just skim through the paper for section and subsection names and figures, tables, charts. This might give you some important clues. When you are a novice you can be Dr. Watson but with enough practice you can be Sherlock Holmes.


What to get from paper?

1. Problem addressed

2. Why to solve this problem?

3. Claims

4. Conclusion

5. Related work

6. Methodology : Assumptions, Algorithm, Data Set, Experiments

7. Possible improvements

8. Connection to other problems

9. What can I do with this knowledge? : The most important


I find multi-pass strategy of Keshav[1] very rewarding. Reading a research paper requires you to be

1. Efficient : You don't have time to read all the research papers at your own leisure.

2. Critical : Not all the research papers published are good (and at times correct as well!).

3. Creative : Reading research paper is just the beginning of the research. How would you build on this acquired knowledge? I will update this documents as and when I change my strategy of reading (and get time to document it).

August 23, 2008

No comments:

Post a Comment