Following notes are taken from a presentation "How to do good research, get it published in SIGKDD and get it cited" by Eamonn Kough
· Must have a problem
· Write before experiments (similar to literature review?)
· Make sure you have real datasets (find the dataset available on internet or ask from peers / authors of other papers solving similar problem or generate dataset)
· Start with simplest idea … convince yourself and then reviewer that this is simples idea
· Don’t’ have solution? Try solutions from other fields …
· Make problem simple by taking assumptions, publish early results and then comeback to original problem by removing assumptions one by one
· Don’t believe everything written in literature is true … don’t trust until you test it
· Make sure problem really exists
· Paper must be written in self explanatory form with self explaining figures so reader should not be forced to think
· Invest time on Title, abstract and introduction: a good introduction with good motivation is half of your success
· Introduction must have following:
o What is the problem? And why this problem is interesting (motivation)?
o Comparison of your approach/methods with others in terms what other methods actually lacking?
o Key components of your approach (basic idea how you got it) including basic assumptions and limitations
o A final paragraph states structure of rest of paper
· Ensure reproducibility, it will help to increase creditability of your work (and so more chances of getting acceptance) and more citation as well.
· Don’t use phrase like attempt to, tries to, seems to …. Write in oblivious way
· States why other approaches can’t work (related work) better is to apply other approaches as well.
· Quote others’ paper:
· Give mean or variance of results instead of actual values
· Make good and clear figures: use direct labeling, color and thickness, minimize the details
· Name columns of table appropriately instead of Method 1 etc. Also don’t use long significant digits
· Always write your paper imagining the most cynical reviewer looking over your shoulder. This reviewer does not particularly like you, does not have a lot of time to spend on your paper, and does not think you are working in an interesting area. But he will listen to reason
Structure of Paper (division in sections is based on my understanding)
o Make a working title
o Introduce the topic and define (informally at this stage) terminology
o Motivation: Emphasize why is the topic important
o Relate to current knowledge: what’s been done
o Indicate the gap: what need’s to be done?
o Formally pose research questions
· Related Work
o Explain any necessary background material.
· Proposed Algorithm/framework
o Introduce formal definitions.
o Introduce your novel algorithm/representation/data structure etc.
o Describe experimental set-up, explain what the experiments will show
· Evaluation and Results
o Describe the datasets
o Summarize results with figures/tables
o Discuss results
o Explain conflicting results, unexpected findings and discrepancies with other research
· Conclusion and Future Work
o State limitations of the study
o State importance of findings
o Announce directions for further research