Database Reference
In-Depth Information
is not enough. You must explain why this previous work does not itself solve your
problem, and what specific limitations of that previous work your approach does
address. Every paper you cite in the previous work section is a fundamental chal-
lenge to the very existence of your project. Your job is to convince a skeptical
reader that the world needs your new thing because it is somehow better than a
particular old thing. Moreover, it's not even enough to just make the case that
yours is different - yours must be
better
. The claims you make must, of course,
be backed up by your validation in a subsequent results section.
A good way to approach the previous work section is that you want to tell
to a story to the reader. Figure out the messages you want to get across to the
reader, in what order, and then use the references to help you tell this story. It
is possible to group the previous work into categories, and to usefully discuss
the limitations of the entire category.
Sweeping Assertions:
A research paper should not contain sweeping
unattributed assertions. You have three choices: cite your source; delete the
assertion from your paper; or explicitly tag the statement as your observation,
your conjecture, or an explanation of your results. In the last case, the assertion
is clearly marked as being part of your research contribution. Be careful with folk
wisdom that “everybody knows”. You could be mistaken, and tracking down the
original sources may change or refine your views. If you cannot find a suitable
source after extensive digging, you have stumbled upon a great topic for a future
paper! You may either validate and extend the conventional wisdom, or show
that it is incorrect.
I Am Utterly Perfect:
No work is perfect. An explicit discussion of the limita-
tions of your work strengthens, rather than weakens, your paper. Papers without
a discussion of limitations, weaknesses, and implications feel unfinished or pre-
liminary. For instance, how large of a dataset can your system handle? Can you
categorize the kinds of datasets for which your technique is suitable and those
for which it is not?
4.3 Results Pitfalls
Several pitfalls on how to validate your claims can occur in the results section
of your paper.
Unfettered by Time:
Do not omit time performance from your writeup, be-
cause it is almost always interesting and worth documenting. The level of detail
at which you should report this result depends on the paper type and the contri-
bution claims. For instance, a very high-level statement like “interactive response
foralldatasetsshownonadesktopPC”maysuceforanevaluationpaperor
a design study paper. However, for a technique paper with a contribution claim
of better performance than previous techniques, detailed comparison timings in
tables or charts would be a better choice.
Search WWH ::
Custom Search