S. Keshav, Paper link
- Make three passes of increasing detail
- All passes have a satisfaction threshold: 5 C's, summarize w/ evidence, point out flaws
The paper itself is a intriguing meta-paper useful for any beginner research. I appreciated the brevity (2 pages) and the algorithmic approach to each pass of the paper. Although the actual mechanics of reading a paper are straightforward, the author takes care to be as expansive as possible. The strategy outlined doesn't specify a field, so researchers from any quantitative field should find the paper useful.
On a more personal level, I had not realized the importance of scanning the references, which is emphasized several times. Citations really are key.
- Quick scan of main details: section titles, references you recognize, gist of paper
- Reviewers will probably only take one pass
- 'Graph abstract' can summarize entire paper in one graph/chart
- Should be able to answer 5 C's: category, context, correctness, contributions, clarity
- Take about an hour
- Ignore details like proofs
- Should be able to summarize paper with evidence to someone else
- Hours for beginners, 1-2 hours for experienced readers
- Reimplement / reprove paper
- Should be able to point out assumptions, missing citations, follow-up questions, and potential issues
- Use search enginer (e.g. Google Scholar) to find 3-5 highly cited papers in field. If survey paper found, you're done
- Otherwise do breadth-first search based on shared references of selected papers, top conferences that referenced authors publish in, and conference proceedings
- How specific should we be about reproving and reimplementing?
- Any examples of academic papers and public feedback? Would be helpful to train the eye.